xref: /linux/drivers/vdpa/vdpa.c (revision d2c9a99135da931377240942d44f3dea104cedb8)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * vDPA bus.
4  *
5  * Copyright (c) 2020, Red Hat. All rights reserved.
6  *     Author: Jason Wang <jasowang@redhat.com>
7  *
8  */
9 
10 #include <linux/module.h>
11 #include <linux/idr.h>
12 #include <linux/slab.h>
13 #include <linux/vdpa.h>
14 #include <uapi/linux/vdpa.h>
15 #include <net/genetlink.h>
16 #include <linux/virtio_ids.h>
17 
18 static LIST_HEAD(mdev_head);
19 /* A global mutex that protects vdpa management device and device level operations. */
20 static DECLARE_RWSEM(vdpa_dev_lock);
21 static DEFINE_IDA(vdpa_index_ida);
22 
vdpa_set_status(struct vdpa_device * vdev,u8 status)23 void vdpa_set_status(struct vdpa_device *vdev, u8 status)
24 {
25 	down_write(&vdev->cf_lock);
26 	vdev->config->set_status(vdev, status);
27 	up_write(&vdev->cf_lock);
28 }
29 EXPORT_SYMBOL(vdpa_set_status);
30 
31 static struct genl_family vdpa_nl_family;
32 
vdpa_dev_probe(struct device * d)33 static int vdpa_dev_probe(struct device *d)
34 {
35 	struct vdpa_device *vdev = dev_to_vdpa(d);
36 	struct vdpa_driver *drv = drv_to_vdpa(vdev->dev.driver);
37 	const struct vdpa_config_ops *ops = vdev->config;
38 	u32 max_num, min_num = 1;
39 	int ret = 0;
40 
41 	d->dma_mask = &d->coherent_dma_mask;
42 	ret = dma_set_mask_and_coherent(d, DMA_BIT_MASK(64));
43 	if (ret)
44 		return ret;
45 
46 	max_num = ops->get_vq_num_max(vdev);
47 	if (ops->get_vq_num_min)
48 		min_num = ops->get_vq_num_min(vdev);
49 	if (max_num < min_num)
50 		return -EINVAL;
51 
52 	if (drv && drv->probe)
53 		ret = drv->probe(vdev);
54 
55 	return ret;
56 }
57 
vdpa_dev_remove(struct device * d)58 static void vdpa_dev_remove(struct device *d)
59 {
60 	struct vdpa_device *vdev = dev_to_vdpa(d);
61 	struct vdpa_driver *drv = drv_to_vdpa(vdev->dev.driver);
62 
63 	if (drv && drv->remove)
64 		drv->remove(vdev);
65 }
66 
vdpa_dev_match(struct device * dev,const struct device_driver * drv)67 static int vdpa_dev_match(struct device *dev, const struct device_driver *drv)
68 {
69 	int ret;
70 
71 	/* Check override first, and if set, only use the named driver */
72 	ret = device_match_driver_override(dev, drv);
73 	if (ret >= 0)
74 		return ret;
75 
76 	/* Currently devices must be supported by all vDPA bus drivers */
77 	return 1;
78 }
79 
80 static const struct bus_type vdpa_bus = {
81 	.name  = "vdpa",
82 	.driver_override = true,
83 	.match = vdpa_dev_match,
84 	.probe = vdpa_dev_probe,
85 	.remove = vdpa_dev_remove,
86 };
87 
vdpa_release_dev(struct device * d)88 static void vdpa_release_dev(struct device *d)
89 {
90 	struct vdpa_device *vdev = dev_to_vdpa(d);
91 	const struct vdpa_config_ops *ops = vdev->config;
92 
93 	if (ops->free)
94 		ops->free(vdev);
95 
96 	ida_free(&vdpa_index_ida, vdev->index);
97 	kfree(vdev);
98 }
99 
100 /**
101  * __vdpa_alloc_device - allocate and initilaize a vDPA device
102  * This allows driver to some prepartion after device is
103  * initialized but before registered.
104  * @parent: the parent device
105  * @config: the bus operations that is supported by this device
106  * @map: the map operations that is supported by this device
107  * @ngroups: number of groups supported by this device
108  * @nas: number of address spaces supported by this device
109  * @size: size of the parent structure that contains private data
110  * @name: name of the vdpa device; optional.
111  * @use_va: indicate whether virtual address must be used by this device
112  *
113  * Driver should use vdpa_alloc_device() wrapper macro instead of
114  * using this directly.
115  *
116  * Return: Returns an error when parent/config/map is not set or fail to get
117  *	   ida.
118  */
__vdpa_alloc_device(struct device * parent,const struct vdpa_config_ops * config,const struct virtio_map_ops * map,unsigned int ngroups,unsigned int nas,size_t size,const char * name,bool use_va)119 struct vdpa_device *__vdpa_alloc_device(struct device *parent,
120 					const struct vdpa_config_ops *config,
121 					const struct virtio_map_ops *map,
122 					unsigned int ngroups, unsigned int nas,
123 					size_t size, const char *name,
124 					bool use_va)
125 {
126 	struct vdpa_device *vdev;
127 	int err = -EINVAL;
128 
129 	if (!config)
130 		goto err;
131 
132 	if (!!config->dma_map != !!config->dma_unmap)
133 		goto err;
134 
135 	/* It should only work for the device that use on-chip IOMMU */
136 	if (use_va && !(config->dma_map || config->set_map))
137 		goto err;
138 
139 	err = -ENOMEM;
140 	vdev = kzalloc(size, GFP_KERNEL);
141 	if (!vdev)
142 		goto err;
143 
144 	err = ida_alloc(&vdpa_index_ida, GFP_KERNEL);
145 	if (err < 0)
146 		goto err_ida;
147 
148 	vdev->dev.bus = &vdpa_bus;
149 	vdev->dev.parent = parent;
150 	vdev->dev.release = vdpa_release_dev;
151 	vdev->index = err;
152 	vdev->config = config;
153 	vdev->map = map;
154 	vdev->features_valid = false;
155 	vdev->use_va = use_va;
156 	vdev->ngroups = ngroups;
157 	vdev->nas = nas;
158 
159 	if (name)
160 		err = dev_set_name(&vdev->dev, "%s", name);
161 	else
162 		err = dev_set_name(&vdev->dev, "vdpa%u", vdev->index);
163 	if (err)
164 		goto err_name;
165 
166 	init_rwsem(&vdev->cf_lock);
167 	device_initialize(&vdev->dev);
168 
169 	return vdev;
170 
171 err_name:
172 	ida_free(&vdpa_index_ida, vdev->index);
173 err_ida:
174 	kfree(vdev);
175 err:
176 	return ERR_PTR(err);
177 }
178 EXPORT_SYMBOL_GPL(__vdpa_alloc_device);
179 
vdpa_name_match(struct device * dev,const void * data)180 static int vdpa_name_match(struct device *dev, const void *data)
181 {
182 	struct vdpa_device *vdev = container_of(dev, struct vdpa_device, dev);
183 
184 	return (strcmp(dev_name(&vdev->dev), data) == 0);
185 }
186 
__vdpa_register_device(struct vdpa_device * vdev,u32 nvqs)187 static int __vdpa_register_device(struct vdpa_device *vdev, u32 nvqs)
188 {
189 	struct device *dev;
190 
191 	vdev->nvqs = nvqs;
192 
193 	lockdep_assert_held(&vdpa_dev_lock);
194 	dev = bus_find_device(&vdpa_bus, NULL, dev_name(&vdev->dev), vdpa_name_match);
195 	if (dev) {
196 		put_device(dev);
197 		return -EEXIST;
198 	}
199 	return device_add(&vdev->dev);
200 }
201 
202 /**
203  * _vdpa_register_device - register a vDPA device with vdpa lock held
204  * Caller must have a succeed call of vdpa_alloc_device() before.
205  * Caller must invoke this routine in the management device dev_add()
206  * callback after setting up valid mgmtdev for this vdpa device.
207  * @vdev: the vdpa device to be registered to vDPA bus
208  * @nvqs: number of virtqueues supported by this device
209  *
210  * Return: Returns an error when fail to add device to vDPA bus
211  */
_vdpa_register_device(struct vdpa_device * vdev,u32 nvqs)212 int _vdpa_register_device(struct vdpa_device *vdev, u32 nvqs)
213 {
214 	if (!vdev->mdev)
215 		return -EINVAL;
216 
217 	return __vdpa_register_device(vdev, nvqs);
218 }
219 EXPORT_SYMBOL_GPL(_vdpa_register_device);
220 
221 /**
222  * vdpa_register_device - register a vDPA device
223  * Callers must have a succeed call of vdpa_alloc_device() before.
224  * @vdev: the vdpa device to be registered to vDPA bus
225  * @nvqs: number of virtqueues supported by this device
226  *
227  * Return: Returns an error when fail to add to vDPA bus
228  */
vdpa_register_device(struct vdpa_device * vdev,u32 nvqs)229 int vdpa_register_device(struct vdpa_device *vdev, u32 nvqs)
230 {
231 	int err;
232 
233 	down_write(&vdpa_dev_lock);
234 	err = __vdpa_register_device(vdev, nvqs);
235 	up_write(&vdpa_dev_lock);
236 	return err;
237 }
238 EXPORT_SYMBOL_GPL(vdpa_register_device);
239 
240 /**
241  * _vdpa_unregister_device - unregister a vDPA device
242  * Caller must invoke this routine as part of management device dev_del()
243  * callback.
244  * @vdev: the vdpa device to be unregisted from vDPA bus
245  */
_vdpa_unregister_device(struct vdpa_device * vdev)246 void _vdpa_unregister_device(struct vdpa_device *vdev)
247 {
248 	lockdep_assert_held(&vdpa_dev_lock);
249 	WARN_ON(!vdev->mdev);
250 	device_unregister(&vdev->dev);
251 }
252 EXPORT_SYMBOL_GPL(_vdpa_unregister_device);
253 
254 /**
255  * vdpa_unregister_device - unregister a vDPA device
256  * @vdev: the vdpa device to be unregisted from vDPA bus
257  */
vdpa_unregister_device(struct vdpa_device * vdev)258 void vdpa_unregister_device(struct vdpa_device *vdev)
259 {
260 	down_write(&vdpa_dev_lock);
261 	device_unregister(&vdev->dev);
262 	up_write(&vdpa_dev_lock);
263 }
264 EXPORT_SYMBOL_GPL(vdpa_unregister_device);
265 
266 /**
267  * __vdpa_register_driver - register a vDPA device driver
268  * @drv: the vdpa device driver to be registered
269  * @owner: module owner of the driver
270  *
271  * Return: Returns an err when fail to do the registration
272  */
__vdpa_register_driver(struct vdpa_driver * drv,struct module * owner)273 int __vdpa_register_driver(struct vdpa_driver *drv, struct module *owner)
274 {
275 	drv->driver.bus = &vdpa_bus;
276 	drv->driver.owner = owner;
277 
278 	return driver_register(&drv->driver);
279 }
280 EXPORT_SYMBOL_GPL(__vdpa_register_driver);
281 
282 /**
283  * vdpa_unregister_driver - unregister a vDPA device driver
284  * @drv: the vdpa device driver to be unregistered
285  */
vdpa_unregister_driver(struct vdpa_driver * drv)286 void vdpa_unregister_driver(struct vdpa_driver *drv)
287 {
288 	driver_unregister(&drv->driver);
289 }
290 EXPORT_SYMBOL_GPL(vdpa_unregister_driver);
291 
292 /**
293  * vdpa_mgmtdev_register - register a vdpa management device
294  *
295  * @mdev: Pointer to vdpa management device
296  * vdpa_mgmtdev_register() register a vdpa management device which supports
297  * vdpa device management.
298  * Return: Returns 0 on success or failure when required callback ops are not
299  *         initialized.
300  */
vdpa_mgmtdev_register(struct vdpa_mgmt_dev * mdev)301 int vdpa_mgmtdev_register(struct vdpa_mgmt_dev *mdev)
302 {
303 	if (!mdev->device || !mdev->ops || !mdev->ops->dev_add || !mdev->ops->dev_del)
304 		return -EINVAL;
305 
306 	INIT_LIST_HEAD(&mdev->list);
307 	down_write(&vdpa_dev_lock);
308 	list_add_tail(&mdev->list, &mdev_head);
309 	up_write(&vdpa_dev_lock);
310 	return 0;
311 }
312 EXPORT_SYMBOL_GPL(vdpa_mgmtdev_register);
313 
vdpa_match_remove(struct device * dev,void * data)314 static int vdpa_match_remove(struct device *dev, void *data)
315 {
316 	struct vdpa_device *vdev = container_of(dev, struct vdpa_device, dev);
317 	struct vdpa_mgmt_dev *mdev = vdev->mdev;
318 
319 	if (mdev == data)
320 		mdev->ops->dev_del(mdev, vdev);
321 	return 0;
322 }
323 
vdpa_mgmtdev_unregister(struct vdpa_mgmt_dev * mdev)324 void vdpa_mgmtdev_unregister(struct vdpa_mgmt_dev *mdev)
325 {
326 	down_write(&vdpa_dev_lock);
327 
328 	list_del(&mdev->list);
329 
330 	/* Filter out all the entries belong to this management device and delete it. */
331 	bus_for_each_dev(&vdpa_bus, NULL, mdev, vdpa_match_remove);
332 
333 	up_write(&vdpa_dev_lock);
334 }
335 EXPORT_SYMBOL_GPL(vdpa_mgmtdev_unregister);
336 
vdpa_get_config_unlocked(struct vdpa_device * vdev,unsigned int offset,void * buf,unsigned int len)337 static void vdpa_get_config_unlocked(struct vdpa_device *vdev,
338 				     unsigned int offset,
339 				     void *buf, unsigned int len)
340 {
341 	const struct vdpa_config_ops *ops = vdev->config;
342 
343 	/*
344 	 * Config accesses aren't supposed to trigger before features are set.
345 	 * If it does happen we assume a legacy guest.
346 	 */
347 	if (!vdev->features_valid)
348 		vdpa_set_features_unlocked(vdev, 0);
349 	ops->get_config(vdev, offset, buf, len);
350 }
351 
352 /**
353  * vdpa_get_config - Get one or more device configuration fields.
354  * @vdev: vdpa device to operate on
355  * @offset: starting byte offset of the field
356  * @buf: buffer pointer to read to
357  * @len: length of the configuration fields in bytes
358  */
vdpa_get_config(struct vdpa_device * vdev,unsigned int offset,void * buf,unsigned int len)359 void vdpa_get_config(struct vdpa_device *vdev, unsigned int offset,
360 		     void *buf, unsigned int len)
361 {
362 	down_read(&vdev->cf_lock);
363 	vdpa_get_config_unlocked(vdev, offset, buf, len);
364 	up_read(&vdev->cf_lock);
365 }
366 EXPORT_SYMBOL_GPL(vdpa_get_config);
367 
368 /**
369  * vdpa_set_config - Set one or more device configuration fields.
370  * @vdev: vdpa device to operate on
371  * @offset: starting byte offset of the field
372  * @buf: buffer pointer to read from
373  * @length: length of the configuration fields in bytes
374  */
vdpa_set_config(struct vdpa_device * vdev,unsigned int offset,const void * buf,unsigned int length)375 void vdpa_set_config(struct vdpa_device *vdev, unsigned int offset,
376 		     const void *buf, unsigned int length)
377 {
378 	down_write(&vdev->cf_lock);
379 	vdev->config->set_config(vdev, offset, buf, length);
380 	up_write(&vdev->cf_lock);
381 }
382 EXPORT_SYMBOL_GPL(vdpa_set_config);
383 
mgmtdev_handle_match(const struct vdpa_mgmt_dev * mdev,const char * busname,const char * devname)384 static bool mgmtdev_handle_match(const struct vdpa_mgmt_dev *mdev,
385 				 const char *busname, const char *devname)
386 {
387 	/* Bus name is optional for simulated management device, so ignore the
388 	 * device with bus if bus attribute is provided.
389 	 */
390 	if ((busname && !mdev->device->bus) || (!busname && mdev->device->bus))
391 		return false;
392 
393 	if (!busname && strcmp(dev_name(mdev->device), devname) == 0)
394 		return true;
395 
396 	if (busname && (strcmp(mdev->device->bus->name, busname) == 0) &&
397 	    (strcmp(dev_name(mdev->device), devname) == 0))
398 		return true;
399 
400 	return false;
401 }
402 
vdpa_mgmtdev_get_from_attr(struct nlattr ** attrs)403 static struct vdpa_mgmt_dev *vdpa_mgmtdev_get_from_attr(struct nlattr **attrs)
404 {
405 	struct vdpa_mgmt_dev *mdev;
406 	const char *busname = NULL;
407 	const char *devname;
408 
409 	if (!attrs[VDPA_ATTR_MGMTDEV_DEV_NAME])
410 		return ERR_PTR(-EINVAL);
411 	devname = nla_data(attrs[VDPA_ATTR_MGMTDEV_DEV_NAME]);
412 	if (attrs[VDPA_ATTR_MGMTDEV_BUS_NAME])
413 		busname = nla_data(attrs[VDPA_ATTR_MGMTDEV_BUS_NAME]);
414 
415 	list_for_each_entry(mdev, &mdev_head, list) {
416 		if (mgmtdev_handle_match(mdev, busname, devname))
417 			return mdev;
418 	}
419 	return ERR_PTR(-ENODEV);
420 }
421 
vdpa_nl_mgmtdev_handle_fill(struct sk_buff * msg,const struct vdpa_mgmt_dev * mdev)422 static int vdpa_nl_mgmtdev_handle_fill(struct sk_buff *msg, const struct vdpa_mgmt_dev *mdev)
423 {
424 	if (mdev->device->bus &&
425 	    nla_put_string(msg, VDPA_ATTR_MGMTDEV_BUS_NAME, mdev->device->bus->name))
426 		return -EMSGSIZE;
427 	if (nla_put_string(msg, VDPA_ATTR_MGMTDEV_DEV_NAME, dev_name(mdev->device)))
428 		return -EMSGSIZE;
429 	return 0;
430 }
431 
vdpa_mgmtdev_get_classes(const struct vdpa_mgmt_dev * mdev,unsigned int * nclasses)432 static u64 vdpa_mgmtdev_get_classes(const struct vdpa_mgmt_dev *mdev,
433 				    unsigned int *nclasses)
434 {
435 	u64 supported_classes = 0;
436 	unsigned int n = 0;
437 
438 	for (int i = 0; mdev->id_table[i].device; i++) {
439 		if (mdev->id_table[i].device > 63)
440 			continue;
441 		supported_classes |= BIT_ULL(mdev->id_table[i].device);
442 		n++;
443 	}
444 	if (nclasses)
445 		*nclasses = n;
446 
447 	return supported_classes;
448 }
449 
vdpa_mgmtdev_fill(const struct vdpa_mgmt_dev * mdev,struct sk_buff * msg,u32 portid,u32 seq,int flags)450 static int vdpa_mgmtdev_fill(const struct vdpa_mgmt_dev *mdev, struct sk_buff *msg,
451 			     u32 portid, u32 seq, int flags)
452 {
453 	void *hdr;
454 	int err;
455 
456 	hdr = genlmsg_put(msg, portid, seq, &vdpa_nl_family, flags, VDPA_CMD_MGMTDEV_NEW);
457 	if (!hdr)
458 		return -EMSGSIZE;
459 	err = vdpa_nl_mgmtdev_handle_fill(msg, mdev);
460 	if (err)
461 		goto msg_err;
462 
463 	if (nla_put_u64_64bit(msg, VDPA_ATTR_MGMTDEV_SUPPORTED_CLASSES,
464 			      vdpa_mgmtdev_get_classes(mdev, NULL),
465 			      VDPA_ATTR_UNSPEC)) {
466 		err = -EMSGSIZE;
467 		goto msg_err;
468 	}
469 	if (nla_put_u32(msg, VDPA_ATTR_DEV_MGMTDEV_MAX_VQS,
470 			mdev->max_supported_vqs)) {
471 		err = -EMSGSIZE;
472 		goto msg_err;
473 	}
474 	if (nla_put_u64_64bit(msg, VDPA_ATTR_DEV_SUPPORTED_FEATURES,
475 			      mdev->supported_features, VDPA_ATTR_PAD)) {
476 		err = -EMSGSIZE;
477 		goto msg_err;
478 	}
479 
480 	genlmsg_end(msg, hdr);
481 	return 0;
482 
483 msg_err:
484 	genlmsg_cancel(msg, hdr);
485 	return err;
486 }
487 
vdpa_nl_cmd_mgmtdev_get_doit(struct sk_buff * skb,struct genl_info * info)488 static int vdpa_nl_cmd_mgmtdev_get_doit(struct sk_buff *skb, struct genl_info *info)
489 {
490 	struct vdpa_mgmt_dev *mdev;
491 	struct sk_buff *msg;
492 	int err;
493 
494 	msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
495 	if (!msg)
496 		return -ENOMEM;
497 
498 	down_read(&vdpa_dev_lock);
499 	mdev = vdpa_mgmtdev_get_from_attr(info->attrs);
500 	if (IS_ERR(mdev)) {
501 		up_read(&vdpa_dev_lock);
502 		NL_SET_ERR_MSG_MOD(info->extack, "Fail to find the specified mgmt device");
503 		err = PTR_ERR(mdev);
504 		goto out;
505 	}
506 
507 	err = vdpa_mgmtdev_fill(mdev, msg, info->snd_portid, info->snd_seq, 0);
508 	up_read(&vdpa_dev_lock);
509 	if (err)
510 		goto out;
511 	err = genlmsg_reply(msg, info);
512 	return err;
513 
514 out:
515 	nlmsg_free(msg);
516 	return err;
517 }
518 
519 static int
vdpa_nl_cmd_mgmtdev_get_dumpit(struct sk_buff * msg,struct netlink_callback * cb)520 vdpa_nl_cmd_mgmtdev_get_dumpit(struct sk_buff *msg, struct netlink_callback *cb)
521 {
522 	struct vdpa_mgmt_dev *mdev;
523 	int start = cb->args[0];
524 	int idx = 0;
525 	int err;
526 
527 	down_read(&vdpa_dev_lock);
528 	list_for_each_entry(mdev, &mdev_head, list) {
529 		if (idx < start) {
530 			idx++;
531 			continue;
532 		}
533 		err = vdpa_mgmtdev_fill(mdev, msg, NETLINK_CB(cb->skb).portid,
534 					cb->nlh->nlmsg_seq, NLM_F_MULTI);
535 		if (err)
536 			goto out;
537 		idx++;
538 	}
539 out:
540 	up_read(&vdpa_dev_lock);
541 	cb->args[0] = idx;
542 	return msg->len;
543 }
544 
545 #define VDPA_DEV_NET_ATTRS_MASK (BIT_ULL(VDPA_ATTR_DEV_NET_CFG_MACADDR) | \
546 				 BIT_ULL(VDPA_ATTR_DEV_NET_CFG_MTU)     | \
547 				 BIT_ULL(VDPA_ATTR_DEV_NET_CFG_MAX_VQP))
548 
549 /*
550  * Bitmask for all per-device features: feature bits VIRTIO_TRANSPORT_F_START
551  * through VIRTIO_TRANSPORT_F_END are unset, i.e. 0xfffffc000fffffff for
552  * all 64bit features. If the features are extended beyond 64 bits, or new
553  * "holes" are reserved for other type of features than per-device, this
554  * macro would have to be updated.
555  */
556 #define VIRTIO_DEVICE_F_MASK (~0ULL << (VIRTIO_TRANSPORT_F_END + 1) | \
557 			      ((1ULL << VIRTIO_TRANSPORT_F_START) - 1))
558 
vdpa_nl_cmd_dev_add_set_doit(struct sk_buff * skb,struct genl_info * info)559 static int vdpa_nl_cmd_dev_add_set_doit(struct sk_buff *skb, struct genl_info *info)
560 {
561 	struct vdpa_dev_set_config config = {};
562 	struct nlattr **nl_attrs = info->attrs;
563 	struct vdpa_mgmt_dev *mdev;
564 	unsigned int ncls = 0;
565 	const u8 *macaddr;
566 	const char *name;
567 	u64 classes;
568 	int err = 0;
569 
570 	if (!info->attrs[VDPA_ATTR_DEV_NAME])
571 		return -EINVAL;
572 
573 	name = nla_data(info->attrs[VDPA_ATTR_DEV_NAME]);
574 
575 	if (nl_attrs[VDPA_ATTR_DEV_NET_CFG_MACADDR]) {
576 		macaddr = nla_data(nl_attrs[VDPA_ATTR_DEV_NET_CFG_MACADDR]);
577 		memcpy(config.net.mac, macaddr, sizeof(config.net.mac));
578 		config.mask |= BIT_ULL(VDPA_ATTR_DEV_NET_CFG_MACADDR);
579 	}
580 	if (nl_attrs[VDPA_ATTR_DEV_NET_CFG_MTU]) {
581 		config.net.mtu =
582 			nla_get_u16(nl_attrs[VDPA_ATTR_DEV_NET_CFG_MTU]);
583 		config.mask |= BIT_ULL(VDPA_ATTR_DEV_NET_CFG_MTU);
584 	}
585 	if (nl_attrs[VDPA_ATTR_DEV_NET_CFG_MAX_VQP]) {
586 		config.net.max_vq_pairs =
587 			nla_get_u16(nl_attrs[VDPA_ATTR_DEV_NET_CFG_MAX_VQP]);
588 		if (!config.net.max_vq_pairs) {
589 			NL_SET_ERR_MSG_MOD(info->extack,
590 					   "At least one pair of VQs is required");
591 			return -EINVAL;
592 		}
593 		config.mask |= BIT_ULL(VDPA_ATTR_DEV_NET_CFG_MAX_VQP);
594 	}
595 	if (nl_attrs[VDPA_ATTR_DEV_FEATURES]) {
596 		u64 missing = 0x0ULL;
597 
598 		config.device_features =
599 			nla_get_u64(nl_attrs[VDPA_ATTR_DEV_FEATURES]);
600 		if (nl_attrs[VDPA_ATTR_DEV_NET_CFG_MACADDR] &&
601 		    !(config.device_features & BIT_ULL(VIRTIO_NET_F_MAC)))
602 			missing |= BIT_ULL(VIRTIO_NET_F_MAC);
603 		if (nl_attrs[VDPA_ATTR_DEV_NET_CFG_MTU] &&
604 		    !(config.device_features & BIT_ULL(VIRTIO_NET_F_MTU)))
605 			missing |= BIT_ULL(VIRTIO_NET_F_MTU);
606 		if (nl_attrs[VDPA_ATTR_DEV_NET_CFG_MAX_VQP] &&
607 		    config.net.max_vq_pairs > 1 &&
608 		    !(config.device_features & BIT_ULL(VIRTIO_NET_F_MQ)))
609 			missing |= BIT_ULL(VIRTIO_NET_F_MQ);
610 		if (missing) {
611 			NL_SET_ERR_MSG_FMT_MOD(info->extack,
612 					       "Missing features 0x%llx for provided attributes",
613 					       missing);
614 			return -EINVAL;
615 		}
616 		config.mask |= BIT_ULL(VDPA_ATTR_DEV_FEATURES);
617 	}
618 
619 	/* Skip checking capability if user didn't prefer to configure any
620 	 * device networking attributes. It is likely that user might have used
621 	 * a device specific method to configure such attributes or using device
622 	 * default attributes.
623 	 */
624 	if ((config.mask & VDPA_DEV_NET_ATTRS_MASK) &&
625 	    !netlink_capable(skb, CAP_NET_ADMIN))
626 		return -EPERM;
627 
628 	down_write(&vdpa_dev_lock);
629 	mdev = vdpa_mgmtdev_get_from_attr(info->attrs);
630 	if (IS_ERR(mdev)) {
631 		NL_SET_ERR_MSG_MOD(info->extack, "Fail to find the specified management device");
632 		err = PTR_ERR(mdev);
633 		goto err;
634 	}
635 
636 	if ((config.mask & mdev->config_attr_mask) != config.mask) {
637 		NL_SET_ERR_MSG_FMT_MOD(info->extack,
638 				       "Some provided attributes are not supported: 0x%llx",
639 				       config.mask & ~mdev->config_attr_mask);
640 		err = -EOPNOTSUPP;
641 		goto err;
642 	}
643 
644 	classes = vdpa_mgmtdev_get_classes(mdev, &ncls);
645 	if (config.mask & VDPA_DEV_NET_ATTRS_MASK &&
646 	    !(classes & BIT_ULL(VIRTIO_ID_NET))) {
647 		NL_SET_ERR_MSG_MOD(info->extack,
648 				   "Network class attributes provided on unsupported management device");
649 		err = -EINVAL;
650 		goto err;
651 	}
652 	if (!(config.mask & VDPA_DEV_NET_ATTRS_MASK) &&
653 	    config.mask & BIT_ULL(VDPA_ATTR_DEV_FEATURES) &&
654 	    classes & BIT_ULL(VIRTIO_ID_NET) && ncls > 1 &&
655 	    config.device_features & VIRTIO_DEVICE_F_MASK) {
656 		NL_SET_ERR_MSG_MOD(info->extack,
657 				   "Management device supports multi-class while device features specified are ambiguous");
658 		err = -EINVAL;
659 		goto err;
660 	}
661 
662 	err = mdev->ops->dev_add(mdev, name, &config);
663 err:
664 	up_write(&vdpa_dev_lock);
665 	return err;
666 }
667 
vdpa_nl_cmd_dev_del_set_doit(struct sk_buff * skb,struct genl_info * info)668 static int vdpa_nl_cmd_dev_del_set_doit(struct sk_buff *skb, struct genl_info *info)
669 {
670 	struct vdpa_mgmt_dev *mdev;
671 	struct vdpa_device *vdev;
672 	struct device *dev;
673 	const char *name;
674 	int err = 0;
675 
676 	if (!info->attrs[VDPA_ATTR_DEV_NAME])
677 		return -EINVAL;
678 	name = nla_data(info->attrs[VDPA_ATTR_DEV_NAME]);
679 
680 	down_write(&vdpa_dev_lock);
681 	dev = bus_find_device(&vdpa_bus, NULL, name, vdpa_name_match);
682 	if (!dev) {
683 		NL_SET_ERR_MSG_MOD(info->extack, "device not found");
684 		err = -ENODEV;
685 		goto dev_err;
686 	}
687 	vdev = container_of(dev, struct vdpa_device, dev);
688 	if (!vdev->mdev) {
689 		NL_SET_ERR_MSG_MOD(info->extack, "Only user created device can be deleted by user");
690 		err = -EINVAL;
691 		goto mdev_err;
692 	}
693 	mdev = vdev->mdev;
694 	mdev->ops->dev_del(mdev, vdev);
695 mdev_err:
696 	put_device(dev);
697 dev_err:
698 	up_write(&vdpa_dev_lock);
699 	return err;
700 }
701 
702 static int
vdpa_dev_fill(struct vdpa_device * vdev,struct sk_buff * msg,u32 portid,u32 seq,int flags,struct netlink_ext_ack * extack)703 vdpa_dev_fill(struct vdpa_device *vdev, struct sk_buff *msg, u32 portid, u32 seq,
704 	      int flags, struct netlink_ext_ack *extack)
705 {
706 	u16 max_vq_size;
707 	u16 min_vq_size = 1;
708 	u32 device_id;
709 	u32 vendor_id;
710 	void *hdr;
711 	int err;
712 
713 	hdr = genlmsg_put(msg, portid, seq, &vdpa_nl_family, flags, VDPA_CMD_DEV_NEW);
714 	if (!hdr)
715 		return -EMSGSIZE;
716 
717 	err = vdpa_nl_mgmtdev_handle_fill(msg, vdev->mdev);
718 	if (err)
719 		goto msg_err;
720 
721 	device_id = vdev->config->get_device_id(vdev);
722 	vendor_id = vdev->config->get_vendor_id(vdev);
723 	max_vq_size = vdev->config->get_vq_num_max(vdev);
724 	if (vdev->config->get_vq_num_min)
725 		min_vq_size = vdev->config->get_vq_num_min(vdev);
726 
727 	err = -EMSGSIZE;
728 	if (nla_put_string(msg, VDPA_ATTR_DEV_NAME, dev_name(&vdev->dev)))
729 		goto msg_err;
730 	if (nla_put_u32(msg, VDPA_ATTR_DEV_ID, device_id))
731 		goto msg_err;
732 	if (nla_put_u32(msg, VDPA_ATTR_DEV_VENDOR_ID, vendor_id))
733 		goto msg_err;
734 	if (nla_put_u32(msg, VDPA_ATTR_DEV_MAX_VQS, vdev->nvqs))
735 		goto msg_err;
736 	if (nla_put_u16(msg, VDPA_ATTR_DEV_MAX_VQ_SIZE, max_vq_size))
737 		goto msg_err;
738 	if (nla_put_u16(msg, VDPA_ATTR_DEV_MIN_VQ_SIZE, min_vq_size))
739 		goto msg_err;
740 
741 	genlmsg_end(msg, hdr);
742 	return 0;
743 
744 msg_err:
745 	genlmsg_cancel(msg, hdr);
746 	return err;
747 }
748 
vdpa_nl_cmd_dev_get_doit(struct sk_buff * skb,struct genl_info * info)749 static int vdpa_nl_cmd_dev_get_doit(struct sk_buff *skb, struct genl_info *info)
750 {
751 	struct vdpa_device *vdev;
752 	struct sk_buff *msg;
753 	const char *devname;
754 	struct device *dev;
755 	int err;
756 
757 	if (!info->attrs[VDPA_ATTR_DEV_NAME])
758 		return -EINVAL;
759 	devname = nla_data(info->attrs[VDPA_ATTR_DEV_NAME]);
760 	msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
761 	if (!msg)
762 		return -ENOMEM;
763 
764 	down_read(&vdpa_dev_lock);
765 	dev = bus_find_device(&vdpa_bus, NULL, devname, vdpa_name_match);
766 	if (!dev) {
767 		NL_SET_ERR_MSG_MOD(info->extack, "device not found");
768 		err = -ENODEV;
769 		goto err;
770 	}
771 	vdev = container_of(dev, struct vdpa_device, dev);
772 	if (!vdev->mdev) {
773 		err = -EINVAL;
774 		goto mdev_err;
775 	}
776 	err = vdpa_dev_fill(vdev, msg, info->snd_portid, info->snd_seq, 0, info->extack);
777 	if (err)
778 		goto mdev_err;
779 
780 	err = genlmsg_reply(msg, info);
781 	put_device(dev);
782 	up_read(&vdpa_dev_lock);
783 	return err;
784 
785 mdev_err:
786 	put_device(dev);
787 err:
788 	up_read(&vdpa_dev_lock);
789 	nlmsg_free(msg);
790 	return err;
791 }
792 
793 struct vdpa_dev_dump_info {
794 	struct sk_buff *msg;
795 	struct netlink_callback *cb;
796 	int start_idx;
797 	int idx;
798 };
799 
vdpa_dev_dump(struct device * dev,void * data)800 static int vdpa_dev_dump(struct device *dev, void *data)
801 {
802 	struct vdpa_device *vdev = container_of(dev, struct vdpa_device, dev);
803 	struct vdpa_dev_dump_info *info = data;
804 	int err;
805 
806 	if (!vdev->mdev)
807 		return 0;
808 	if (info->idx < info->start_idx) {
809 		info->idx++;
810 		return 0;
811 	}
812 	err = vdpa_dev_fill(vdev, info->msg, NETLINK_CB(info->cb->skb).portid,
813 			    info->cb->nlh->nlmsg_seq, NLM_F_MULTI, info->cb->extack);
814 	if (err)
815 		return err;
816 
817 	info->idx++;
818 	return 0;
819 }
820 
vdpa_nl_cmd_dev_get_dumpit(struct sk_buff * msg,struct netlink_callback * cb)821 static int vdpa_nl_cmd_dev_get_dumpit(struct sk_buff *msg, struct netlink_callback *cb)
822 {
823 	struct vdpa_dev_dump_info info;
824 
825 	info.msg = msg;
826 	info.cb = cb;
827 	info.start_idx = cb->args[0];
828 	info.idx = 0;
829 
830 	down_read(&vdpa_dev_lock);
831 	bus_for_each_dev(&vdpa_bus, NULL, &info, vdpa_dev_dump);
832 	up_read(&vdpa_dev_lock);
833 	cb->args[0] = info.idx;
834 	return msg->len;
835 }
836 
vdpa_dev_net_mq_config_fill(struct sk_buff * msg,u64 features,const struct virtio_net_config * config)837 static int vdpa_dev_net_mq_config_fill(struct sk_buff *msg, u64 features,
838 				       const struct virtio_net_config *config)
839 {
840 	u16 val_u16;
841 
842 	if ((features & BIT_ULL(VIRTIO_NET_F_MQ)) == 0 &&
843 	    (features & BIT_ULL(VIRTIO_NET_F_RSS)) == 0)
844 		return 0;
845 
846 	val_u16 = __virtio16_to_cpu(true, config->max_virtqueue_pairs);
847 
848 	return nla_put_u16(msg, VDPA_ATTR_DEV_NET_CFG_MAX_VQP, val_u16);
849 }
850 
vdpa_dev_net_mtu_config_fill(struct sk_buff * msg,u64 features,const struct virtio_net_config * config)851 static int vdpa_dev_net_mtu_config_fill(struct sk_buff *msg, u64 features,
852 					const struct virtio_net_config *config)
853 {
854 	u16 val_u16;
855 
856 	if ((features & BIT_ULL(VIRTIO_NET_F_MTU)) == 0)
857 		return 0;
858 
859 	val_u16 = __virtio16_to_cpu(true, config->mtu);
860 
861 	return nla_put_u16(msg, VDPA_ATTR_DEV_NET_CFG_MTU, val_u16);
862 }
863 
vdpa_dev_net_mac_config_fill(struct sk_buff * msg,u64 features,const struct virtio_net_config * config)864 static int vdpa_dev_net_mac_config_fill(struct sk_buff *msg, u64 features,
865 					const struct virtio_net_config *config)
866 {
867 	if ((features & BIT_ULL(VIRTIO_NET_F_MAC)) == 0)
868 		return 0;
869 
870 	return  nla_put(msg, VDPA_ATTR_DEV_NET_CFG_MACADDR,
871 			sizeof(config->mac), config->mac);
872 }
873 
vdpa_dev_net_status_config_fill(struct sk_buff * msg,u64 features,const struct virtio_net_config * config)874 static int vdpa_dev_net_status_config_fill(struct sk_buff *msg, u64 features,
875 					   const struct virtio_net_config *config)
876 {
877 	u16 val_u16;
878 
879 	if ((features & BIT_ULL(VIRTIO_NET_F_STATUS)) == 0)
880 		return 0;
881 
882 	val_u16 = __virtio16_to_cpu(true, config->status);
883 	return nla_put_u16(msg, VDPA_ATTR_DEV_NET_STATUS, val_u16);
884 }
885 
vdpa_dev_net_config_fill(struct vdpa_device * vdev,struct sk_buff * msg)886 static int vdpa_dev_net_config_fill(struct vdpa_device *vdev, struct sk_buff *msg)
887 {
888 	struct virtio_net_config config = {};
889 	u64 features_device;
890 
891 	vdev->config->get_config(vdev, 0, &config, sizeof(config));
892 
893 	features_device = vdev->config->get_device_features(vdev);
894 
895 	if (nla_put_u64_64bit(msg, VDPA_ATTR_DEV_FEATURES, features_device,
896 			      VDPA_ATTR_PAD))
897 		return -EMSGSIZE;
898 
899 	if (vdpa_dev_net_mtu_config_fill(msg, features_device, &config))
900 		return -EMSGSIZE;
901 
902 	if (vdpa_dev_net_mac_config_fill(msg, features_device, &config))
903 		return -EMSGSIZE;
904 
905 	if (vdpa_dev_net_status_config_fill(msg, features_device, &config))
906 		return -EMSGSIZE;
907 
908 	return vdpa_dev_net_mq_config_fill(msg, features_device, &config);
909 }
910 
911 static int
vdpa_dev_blk_capacity_config_fill(struct sk_buff * msg,const struct virtio_blk_config * config)912 vdpa_dev_blk_capacity_config_fill(struct sk_buff *msg,
913 				  const struct virtio_blk_config *config)
914 {
915 	u64 val_u64;
916 
917 	val_u64 = __virtio64_to_cpu(true, config->capacity);
918 
919 	return nla_put_u64_64bit(msg, VDPA_ATTR_DEV_BLK_CFG_CAPACITY,
920 				 val_u64, VDPA_ATTR_PAD);
921 }
922 
923 static int
vdpa_dev_blk_seg_size_config_fill(struct sk_buff * msg,u64 features,const struct virtio_blk_config * config)924 vdpa_dev_blk_seg_size_config_fill(struct sk_buff *msg, u64 features,
925 				  const struct virtio_blk_config *config)
926 {
927 	u32 val_u32;
928 
929 	if ((features & BIT_ULL(VIRTIO_BLK_F_SIZE_MAX)) == 0)
930 		return 0;
931 
932 	val_u32 = __virtio32_to_cpu(true, config->size_max);
933 
934 	return nla_put_u32(msg, VDPA_ATTR_DEV_BLK_CFG_SIZE_MAX, val_u32);
935 }
936 
937 /* fill the block size*/
938 static int
vdpa_dev_blk_block_size_config_fill(struct sk_buff * msg,u64 features,const struct virtio_blk_config * config)939 vdpa_dev_blk_block_size_config_fill(struct sk_buff *msg, u64 features,
940 				    const struct virtio_blk_config *config)
941 {
942 	u32 val_u32;
943 
944 	if ((features & BIT_ULL(VIRTIO_BLK_F_BLK_SIZE)) == 0)
945 		return 0;
946 
947 	val_u32 = __virtio32_to_cpu(true, config->blk_size);
948 
949 	return nla_put_u32(msg, VDPA_ATTR_DEV_BLK_CFG_BLK_SIZE, val_u32);
950 }
951 
952 static int
vdpa_dev_blk_seg_max_config_fill(struct sk_buff * msg,u64 features,const struct virtio_blk_config * config)953 vdpa_dev_blk_seg_max_config_fill(struct sk_buff *msg, u64 features,
954 				 const struct virtio_blk_config *config)
955 {
956 	u32 val_u32;
957 
958 	if ((features & BIT_ULL(VIRTIO_BLK_F_SEG_MAX)) == 0)
959 		return 0;
960 
961 	val_u32 = __virtio32_to_cpu(true, config->seg_max);
962 
963 	return nla_put_u32(msg, VDPA_ATTR_DEV_BLK_CFG_SEG_MAX, val_u32);
964 }
965 
vdpa_dev_blk_mq_config_fill(struct sk_buff * msg,u64 features,const struct virtio_blk_config * config)966 static int vdpa_dev_blk_mq_config_fill(struct sk_buff *msg, u64 features,
967 				       const struct virtio_blk_config *config)
968 {
969 	u16 val_u16;
970 
971 	if ((features & BIT_ULL(VIRTIO_BLK_F_MQ)) == 0)
972 		return 0;
973 
974 	val_u16 = __virtio16_to_cpu(true, config->num_queues);
975 
976 	return nla_put_u16(msg, VDPA_ATTR_DEV_BLK_CFG_NUM_QUEUES, val_u16);
977 }
978 
vdpa_dev_blk_topology_config_fill(struct sk_buff * msg,u64 features,const struct virtio_blk_config * config)979 static int vdpa_dev_blk_topology_config_fill(struct sk_buff *msg, u64 features,
980 				       const struct virtio_blk_config *config)
981 {
982 	u16 min_io_size;
983 	u32 opt_io_size;
984 
985 	if ((features & BIT_ULL(VIRTIO_BLK_F_TOPOLOGY)) == 0)
986 		return 0;
987 
988 	min_io_size = __virtio16_to_cpu(true, config->min_io_size);
989 	opt_io_size = __virtio32_to_cpu(true, config->opt_io_size);
990 
991 	if (nla_put_u8(msg, VDPA_ATTR_DEV_BLK_CFG_PHY_BLK_EXP,
992 	    config->physical_block_exp))
993 		return -EMSGSIZE;
994 
995 	if (nla_put_u8(msg, VDPA_ATTR_DEV_BLK_CFG_ALIGN_OFFSET,
996 	    config->alignment_offset))
997 		return -EMSGSIZE;
998 
999 	if (nla_put_u16(msg, VDPA_ATTR_DEV_BLK_CFG_MIN_IO_SIZE, min_io_size))
1000 		return -EMSGSIZE;
1001 
1002 	if (nla_put_u32(msg, VDPA_ATTR_DEV_BLK_CFG_OPT_IO_SIZE, opt_io_size))
1003 		return -EMSGSIZE;
1004 
1005 	return 0;
1006 }
1007 
vdpa_dev_blk_discard_config_fill(struct sk_buff * msg,u64 features,const struct virtio_blk_config * config)1008 static int vdpa_dev_blk_discard_config_fill(struct sk_buff *msg, u64 features,
1009 				       const struct virtio_blk_config *config)
1010 {
1011 	u32 val_u32;
1012 
1013 	if ((features & BIT_ULL(VIRTIO_BLK_F_DISCARD)) == 0)
1014 		return 0;
1015 
1016 	val_u32 = __virtio32_to_cpu(true, config->max_discard_sectors);
1017 	if (nla_put_u32(msg, VDPA_ATTR_DEV_BLK_CFG_MAX_DISCARD_SEC, val_u32))
1018 		return -EMSGSIZE;
1019 
1020 	val_u32 = __virtio32_to_cpu(true, config->max_discard_seg);
1021 	if (nla_put_u32(msg, VDPA_ATTR_DEV_BLK_CFG_MAX_DISCARD_SEG, val_u32))
1022 		return -EMSGSIZE;
1023 
1024 	val_u32 = __virtio32_to_cpu(true, config->discard_sector_alignment);
1025 	if (nla_put_u32(msg, VDPA_ATTR_DEV_BLK_CFG_DISCARD_SEC_ALIGN, val_u32))
1026 		return -EMSGSIZE;
1027 
1028 	return 0;
1029 }
1030 
1031 static int
vdpa_dev_blk_write_zeroes_config_fill(struct sk_buff * msg,u64 features,const struct virtio_blk_config * config)1032 vdpa_dev_blk_write_zeroes_config_fill(struct sk_buff *msg, u64 features,
1033 				     const struct virtio_blk_config *config)
1034 {
1035 	u32 val_u32;
1036 
1037 	if ((features & BIT_ULL(VIRTIO_BLK_F_WRITE_ZEROES)) == 0)
1038 		return 0;
1039 
1040 	val_u32 = __virtio32_to_cpu(true, config->max_write_zeroes_sectors);
1041 	if (nla_put_u32(msg, VDPA_ATTR_DEV_BLK_CFG_MAX_WRITE_ZEROES_SEC, val_u32))
1042 		return -EMSGSIZE;
1043 
1044 	val_u32 = __virtio32_to_cpu(true, config->max_write_zeroes_seg);
1045 	if (nla_put_u32(msg, VDPA_ATTR_DEV_BLK_CFG_MAX_WRITE_ZEROES_SEG, val_u32))
1046 		return -EMSGSIZE;
1047 
1048 	return 0;
1049 }
1050 
vdpa_dev_blk_ro_config_fill(struct sk_buff * msg,u64 features)1051 static int vdpa_dev_blk_ro_config_fill(struct sk_buff *msg, u64 features)
1052 {
1053 	u8 ro;
1054 
1055 	ro = ((features & BIT_ULL(VIRTIO_BLK_F_RO)) == 0) ? 0 : 1;
1056 	if (nla_put_u8(msg, VDPA_ATTR_DEV_BLK_READ_ONLY, ro))
1057 		return -EMSGSIZE;
1058 
1059 	return 0;
1060 }
1061 
vdpa_dev_blk_flush_config_fill(struct sk_buff * msg,u64 features)1062 static int vdpa_dev_blk_flush_config_fill(struct sk_buff *msg, u64 features)
1063 {
1064 	u8 flush;
1065 
1066 	flush = ((features & BIT_ULL(VIRTIO_BLK_F_FLUSH)) == 0) ? 0 : 1;
1067 	if (nla_put_u8(msg, VDPA_ATTR_DEV_BLK_FLUSH, flush))
1068 		return -EMSGSIZE;
1069 
1070 	return 0;
1071 }
1072 
vdpa_dev_blk_config_fill(struct vdpa_device * vdev,struct sk_buff * msg)1073 static int vdpa_dev_blk_config_fill(struct vdpa_device *vdev,
1074 				    struct sk_buff *msg)
1075 {
1076 	struct virtio_blk_config config = {};
1077 	u64 features_device;
1078 
1079 	vdev->config->get_config(vdev, 0, &config, sizeof(config));
1080 
1081 	features_device = vdev->config->get_device_features(vdev);
1082 
1083 	if (nla_put_u64_64bit(msg, VDPA_ATTR_DEV_FEATURES, features_device,
1084 			      VDPA_ATTR_PAD))
1085 		return -EMSGSIZE;
1086 
1087 	if (vdpa_dev_blk_capacity_config_fill(msg, &config))
1088 		return -EMSGSIZE;
1089 
1090 	if (vdpa_dev_blk_seg_size_config_fill(msg, features_device, &config))
1091 		return -EMSGSIZE;
1092 
1093 	if (vdpa_dev_blk_block_size_config_fill(msg, features_device, &config))
1094 		return -EMSGSIZE;
1095 
1096 	if (vdpa_dev_blk_seg_max_config_fill(msg, features_device, &config))
1097 		return -EMSGSIZE;
1098 
1099 	if (vdpa_dev_blk_mq_config_fill(msg, features_device, &config))
1100 		return -EMSGSIZE;
1101 
1102 	if (vdpa_dev_blk_topology_config_fill(msg, features_device, &config))
1103 		return -EMSGSIZE;
1104 
1105 	if (vdpa_dev_blk_discard_config_fill(msg, features_device, &config))
1106 		return -EMSGSIZE;
1107 
1108 	if (vdpa_dev_blk_write_zeroes_config_fill(msg, features_device, &config))
1109 		return -EMSGSIZE;
1110 
1111 	if (vdpa_dev_blk_ro_config_fill(msg, features_device))
1112 		return -EMSGSIZE;
1113 
1114 	if (vdpa_dev_blk_flush_config_fill(msg, features_device))
1115 		return -EMSGSIZE;
1116 
1117 	return 0;
1118 }
1119 
1120 static int
vdpa_dev_config_fill(struct vdpa_device * vdev,struct sk_buff * msg,u32 portid,u32 seq,int flags,struct netlink_ext_ack * extack)1121 vdpa_dev_config_fill(struct vdpa_device *vdev, struct sk_buff *msg, u32 portid, u32 seq,
1122 		     int flags, struct netlink_ext_ack *extack)
1123 {
1124 	u64 features_driver;
1125 	u8 status = 0;
1126 	u32 device_id;
1127 	void *hdr;
1128 	int err;
1129 
1130 	down_read(&vdev->cf_lock);
1131 	hdr = genlmsg_put(msg, portid, seq, &vdpa_nl_family, flags,
1132 			  VDPA_CMD_DEV_CONFIG_GET);
1133 	if (!hdr) {
1134 		err = -EMSGSIZE;
1135 		goto out;
1136 	}
1137 
1138 	if (nla_put_string(msg, VDPA_ATTR_DEV_NAME, dev_name(&vdev->dev))) {
1139 		err = -EMSGSIZE;
1140 		goto msg_err;
1141 	}
1142 
1143 	device_id = vdev->config->get_device_id(vdev);
1144 	if (nla_put_u32(msg, VDPA_ATTR_DEV_ID, device_id)) {
1145 		err = -EMSGSIZE;
1146 		goto msg_err;
1147 	}
1148 
1149 	/* only read driver features after the feature negotiation is done */
1150 	status = vdev->config->get_status(vdev);
1151 	if (status & VIRTIO_CONFIG_S_FEATURES_OK) {
1152 		features_driver = vdev->config->get_driver_features(vdev);
1153 		if (nla_put_u64_64bit(msg, VDPA_ATTR_DEV_NEGOTIATED_FEATURES, features_driver,
1154 				      VDPA_ATTR_PAD)) {
1155 			err = -EMSGSIZE;
1156 			goto msg_err;
1157 		}
1158 	}
1159 
1160 	switch (device_id) {
1161 	case VIRTIO_ID_NET:
1162 		err = vdpa_dev_net_config_fill(vdev, msg);
1163 		break;
1164 	case VIRTIO_ID_BLOCK:
1165 		err = vdpa_dev_blk_config_fill(vdev, msg);
1166 		break;
1167 	default:
1168 		err = -EOPNOTSUPP;
1169 		break;
1170 	}
1171 	if (err)
1172 		goto msg_err;
1173 
1174 	up_read(&vdev->cf_lock);
1175 	genlmsg_end(msg, hdr);
1176 	return 0;
1177 
1178 msg_err:
1179 	genlmsg_cancel(msg, hdr);
1180 out:
1181 	up_read(&vdev->cf_lock);
1182 	return err;
1183 }
1184 
vdpa_fill_stats_rec(struct vdpa_device * vdev,struct sk_buff * msg,struct genl_info * info,u32 index)1185 static int vdpa_fill_stats_rec(struct vdpa_device *vdev, struct sk_buff *msg,
1186 			       struct genl_info *info, u32 index)
1187 {
1188 	struct virtio_net_config config = {};
1189 	u64 features;
1190 	u8 status;
1191 	int err;
1192 
1193 	status = vdev->config->get_status(vdev);
1194 	if (!(status & VIRTIO_CONFIG_S_FEATURES_OK)) {
1195 		NL_SET_ERR_MSG_MOD(info->extack, "feature negotiation not complete");
1196 		return -EAGAIN;
1197 	}
1198 	vdpa_get_config_unlocked(vdev, 0, &config, sizeof(config));
1199 
1200 	features = vdev->config->get_driver_features(vdev);
1201 	if (nla_put_u64_64bit(msg, VDPA_ATTR_DEV_NEGOTIATED_FEATURES,
1202 			      features, VDPA_ATTR_PAD))
1203 		return -EMSGSIZE;
1204 
1205 	err = vdpa_dev_net_mq_config_fill(msg, features, &config);
1206 	if (err)
1207 		return err;
1208 
1209 	if (nla_put_u32(msg, VDPA_ATTR_DEV_QUEUE_INDEX, index))
1210 		return -EMSGSIZE;
1211 
1212 	err = vdev->config->get_vendor_vq_stats(vdev, index, msg, info->extack);
1213 	if (err)
1214 		return err;
1215 
1216 	return 0;
1217 }
1218 
vendor_stats_fill(struct vdpa_device * vdev,struct sk_buff * msg,struct genl_info * info,u32 index)1219 static int vendor_stats_fill(struct vdpa_device *vdev, struct sk_buff *msg,
1220 			     struct genl_info *info, u32 index)
1221 {
1222 	int err;
1223 
1224 	down_read(&vdev->cf_lock);
1225 	if (!vdev->config->get_vendor_vq_stats) {
1226 		err = -EOPNOTSUPP;
1227 		goto out;
1228 	}
1229 
1230 	err = vdpa_fill_stats_rec(vdev, msg, info, index);
1231 out:
1232 	up_read(&vdev->cf_lock);
1233 	return err;
1234 }
1235 
vdpa_dev_vendor_stats_fill(struct vdpa_device * vdev,struct sk_buff * msg,struct genl_info * info,u32 index)1236 static int vdpa_dev_vendor_stats_fill(struct vdpa_device *vdev,
1237 				      struct sk_buff *msg,
1238 				      struct genl_info *info, u32 index)
1239 {
1240 	u32 device_id;
1241 	void *hdr;
1242 	int err;
1243 	u32 portid = info->snd_portid;
1244 	u32 seq = info->snd_seq;
1245 	u32 flags = 0;
1246 
1247 	hdr = genlmsg_put(msg, portid, seq, &vdpa_nl_family, flags,
1248 			  VDPA_CMD_DEV_VSTATS_GET);
1249 	if (!hdr)
1250 		return -EMSGSIZE;
1251 
1252 	if (nla_put_string(msg, VDPA_ATTR_DEV_NAME, dev_name(&vdev->dev))) {
1253 		err = -EMSGSIZE;
1254 		goto undo_msg;
1255 	}
1256 
1257 	device_id = vdev->config->get_device_id(vdev);
1258 	if (nla_put_u32(msg, VDPA_ATTR_DEV_ID, device_id)) {
1259 		err = -EMSGSIZE;
1260 		goto undo_msg;
1261 	}
1262 
1263 	switch (device_id) {
1264 	case VIRTIO_ID_NET:
1265 		if (index > VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MAX) {
1266 			NL_SET_ERR_MSG_MOD(info->extack, "queue index exceeds max value");
1267 			err = -ERANGE;
1268 			break;
1269 		}
1270 
1271 		err = vendor_stats_fill(vdev, msg, info, index);
1272 		break;
1273 	default:
1274 		err = -EOPNOTSUPP;
1275 		break;
1276 	}
1277 	genlmsg_end(msg, hdr);
1278 
1279 	return err;
1280 
1281 undo_msg:
1282 	genlmsg_cancel(msg, hdr);
1283 	return err;
1284 }
1285 
vdpa_nl_cmd_dev_config_get_doit(struct sk_buff * skb,struct genl_info * info)1286 static int vdpa_nl_cmd_dev_config_get_doit(struct sk_buff *skb, struct genl_info *info)
1287 {
1288 	struct vdpa_device *vdev;
1289 	struct sk_buff *msg;
1290 	const char *devname;
1291 	struct device *dev;
1292 	int err;
1293 
1294 	if (!info->attrs[VDPA_ATTR_DEV_NAME])
1295 		return -EINVAL;
1296 	devname = nla_data(info->attrs[VDPA_ATTR_DEV_NAME]);
1297 	msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1298 	if (!msg)
1299 		return -ENOMEM;
1300 
1301 	down_read(&vdpa_dev_lock);
1302 	dev = bus_find_device(&vdpa_bus, NULL, devname, vdpa_name_match);
1303 	if (!dev) {
1304 		NL_SET_ERR_MSG_MOD(info->extack, "device not found");
1305 		err = -ENODEV;
1306 		goto dev_err;
1307 	}
1308 	vdev = container_of(dev, struct vdpa_device, dev);
1309 	if (!vdev->mdev) {
1310 		NL_SET_ERR_MSG_MOD(info->extack, "unmanaged vdpa device");
1311 		err = -EINVAL;
1312 		goto mdev_err;
1313 	}
1314 	err = vdpa_dev_config_fill(vdev, msg, info->snd_portid, info->snd_seq,
1315 				   0, info->extack);
1316 	if (!err)
1317 		err = genlmsg_reply(msg, info);
1318 
1319 mdev_err:
1320 	put_device(dev);
1321 dev_err:
1322 	up_read(&vdpa_dev_lock);
1323 	if (err)
1324 		nlmsg_free(msg);
1325 	return err;
1326 }
1327 
vdpa_dev_net_device_attr_set(struct vdpa_device * vdev,struct genl_info * info)1328 static int vdpa_dev_net_device_attr_set(struct vdpa_device *vdev,
1329 					struct genl_info *info)
1330 {
1331 	struct vdpa_dev_set_config set_config = {};
1332 	struct vdpa_mgmt_dev *mdev = vdev->mdev;
1333 	struct nlattr **nl_attrs = info->attrs;
1334 	const u8 *macaddr;
1335 	int err = -EOPNOTSUPP;
1336 
1337 	down_write(&vdev->cf_lock);
1338 	if (nl_attrs[VDPA_ATTR_DEV_NET_CFG_MACADDR]) {
1339 		set_config.mask |= BIT_ULL(VDPA_ATTR_DEV_NET_CFG_MACADDR);
1340 		macaddr = nla_data(nl_attrs[VDPA_ATTR_DEV_NET_CFG_MACADDR]);
1341 
1342 		if (is_valid_ether_addr(macaddr)) {
1343 			ether_addr_copy(set_config.net.mac, macaddr);
1344 			if (mdev->ops->dev_set_attr) {
1345 				err = mdev->ops->dev_set_attr(mdev, vdev,
1346 							      &set_config);
1347 			} else {
1348 				NL_SET_ERR_MSG_FMT_MOD(info->extack,
1349 						       "Operation not supported by the device.");
1350 			}
1351 		} else {
1352 			NL_SET_ERR_MSG_FMT_MOD(info->extack,
1353 					       "Invalid MAC address");
1354 		}
1355 	}
1356 	up_write(&vdev->cf_lock);
1357 	return err;
1358 }
1359 
vdpa_nl_cmd_dev_attr_set_doit(struct sk_buff * skb,struct genl_info * info)1360 static int vdpa_nl_cmd_dev_attr_set_doit(struct sk_buff *skb,
1361 					 struct genl_info *info)
1362 {
1363 	struct vdpa_device *vdev;
1364 	struct device *dev;
1365 	const char *name;
1366 	u64 classes;
1367 	int err = 0;
1368 
1369 	if (!info->attrs[VDPA_ATTR_DEV_NAME])
1370 		return -EINVAL;
1371 
1372 	name = nla_data(info->attrs[VDPA_ATTR_DEV_NAME]);
1373 
1374 	down_write(&vdpa_dev_lock);
1375 	dev = bus_find_device(&vdpa_bus, NULL, name, vdpa_name_match);
1376 	if (!dev) {
1377 		NL_SET_ERR_MSG_MOD(info->extack, "device not found");
1378 		err = -ENODEV;
1379 		goto dev_err;
1380 	}
1381 	vdev = container_of(dev, struct vdpa_device, dev);
1382 	if (!vdev->mdev) {
1383 		NL_SET_ERR_MSG_MOD(info->extack, "unmanaged vdpa device");
1384 		err = -EINVAL;
1385 		goto mdev_err;
1386 	}
1387 	classes = vdpa_mgmtdev_get_classes(vdev->mdev, NULL);
1388 	if (classes & BIT_ULL(VIRTIO_ID_NET)) {
1389 		err = vdpa_dev_net_device_attr_set(vdev, info);
1390 	} else {
1391 		NL_SET_ERR_MSG_FMT_MOD(info->extack, "%s device not supported",
1392 				       name);
1393 	}
1394 
1395 mdev_err:
1396 	put_device(dev);
1397 dev_err:
1398 	up_write(&vdpa_dev_lock);
1399 	return err;
1400 }
1401 
vdpa_dev_config_dump(struct device * dev,void * data)1402 static int vdpa_dev_config_dump(struct device *dev, void *data)
1403 {
1404 	struct vdpa_device *vdev = container_of(dev, struct vdpa_device, dev);
1405 	struct vdpa_dev_dump_info *info = data;
1406 	int err;
1407 
1408 	if (!vdev->mdev)
1409 		return 0;
1410 	if (info->idx < info->start_idx) {
1411 		info->idx++;
1412 		return 0;
1413 	}
1414 	err = vdpa_dev_config_fill(vdev, info->msg, NETLINK_CB(info->cb->skb).portid,
1415 				   info->cb->nlh->nlmsg_seq, NLM_F_MULTI,
1416 				   info->cb->extack);
1417 	if (err)
1418 		return err;
1419 
1420 	info->idx++;
1421 	return 0;
1422 }
1423 
1424 static int
vdpa_nl_cmd_dev_config_get_dumpit(struct sk_buff * msg,struct netlink_callback * cb)1425 vdpa_nl_cmd_dev_config_get_dumpit(struct sk_buff *msg, struct netlink_callback *cb)
1426 {
1427 	struct vdpa_dev_dump_info info;
1428 
1429 	info.msg = msg;
1430 	info.cb = cb;
1431 	info.start_idx = cb->args[0];
1432 	info.idx = 0;
1433 
1434 	down_read(&vdpa_dev_lock);
1435 	bus_for_each_dev(&vdpa_bus, NULL, &info, vdpa_dev_config_dump);
1436 	up_read(&vdpa_dev_lock);
1437 	cb->args[0] = info.idx;
1438 	return msg->len;
1439 }
1440 
vdpa_nl_cmd_dev_stats_get_doit(struct sk_buff * skb,struct genl_info * info)1441 static int vdpa_nl_cmd_dev_stats_get_doit(struct sk_buff *skb,
1442 					  struct genl_info *info)
1443 {
1444 	struct vdpa_device *vdev;
1445 	struct sk_buff *msg;
1446 	const char *devname;
1447 	struct device *dev;
1448 	u32 index;
1449 	int err;
1450 
1451 	if (!info->attrs[VDPA_ATTR_DEV_NAME])
1452 		return -EINVAL;
1453 
1454 	if (!info->attrs[VDPA_ATTR_DEV_QUEUE_INDEX])
1455 		return -EINVAL;
1456 
1457 	devname = nla_data(info->attrs[VDPA_ATTR_DEV_NAME]);
1458 	msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1459 	if (!msg)
1460 		return -ENOMEM;
1461 
1462 	index = nla_get_u32(info->attrs[VDPA_ATTR_DEV_QUEUE_INDEX]);
1463 	down_read(&vdpa_dev_lock);
1464 	dev = bus_find_device(&vdpa_bus, NULL, devname, vdpa_name_match);
1465 	if (!dev) {
1466 		NL_SET_ERR_MSG_MOD(info->extack, "device not found");
1467 		err = -ENODEV;
1468 		goto dev_err;
1469 	}
1470 	vdev = container_of(dev, struct vdpa_device, dev);
1471 	if (!vdev->mdev) {
1472 		NL_SET_ERR_MSG_MOD(info->extack, "unmanaged vdpa device");
1473 		err = -EINVAL;
1474 		goto mdev_err;
1475 	}
1476 	err = vdpa_dev_vendor_stats_fill(vdev, msg, info, index);
1477 	if (err)
1478 		goto mdev_err;
1479 
1480 	err = genlmsg_reply(msg, info);
1481 
1482 	put_device(dev);
1483 	up_read(&vdpa_dev_lock);
1484 
1485 	return err;
1486 
1487 mdev_err:
1488 	put_device(dev);
1489 dev_err:
1490 	nlmsg_free(msg);
1491 	up_read(&vdpa_dev_lock);
1492 	return err;
1493 }
1494 
1495 static const struct nla_policy vdpa_nl_policy[VDPA_ATTR_MAX + 1] = {
1496 	[VDPA_ATTR_MGMTDEV_BUS_NAME] = { .type = NLA_NUL_STRING },
1497 	[VDPA_ATTR_MGMTDEV_DEV_NAME] = { .type = NLA_STRING },
1498 	[VDPA_ATTR_DEV_NAME] = { .type = NLA_STRING },
1499 	[VDPA_ATTR_DEV_NET_CFG_MACADDR] = NLA_POLICY_ETH_ADDR,
1500 	[VDPA_ATTR_DEV_NET_CFG_MAX_VQP] = { .type = NLA_U16 },
1501 	/* virtio spec 1.1 section 5.1.4.1 for valid MTU range */
1502 	[VDPA_ATTR_DEV_NET_CFG_MTU] = NLA_POLICY_MIN(NLA_U16, 68),
1503 	[VDPA_ATTR_DEV_QUEUE_INDEX] = { .type = NLA_U32 },
1504 	[VDPA_ATTR_DEV_FEATURES] = { .type = NLA_U64 },
1505 };
1506 
1507 static const struct genl_ops vdpa_nl_ops[] = {
1508 	{
1509 		.cmd = VDPA_CMD_MGMTDEV_GET,
1510 		.doit = vdpa_nl_cmd_mgmtdev_get_doit,
1511 		.dumpit = vdpa_nl_cmd_mgmtdev_get_dumpit,
1512 	},
1513 	{
1514 		.cmd = VDPA_CMD_DEV_NEW,
1515 		.doit = vdpa_nl_cmd_dev_add_set_doit,
1516 		.flags = GENL_ADMIN_PERM,
1517 	},
1518 	{
1519 		.cmd = VDPA_CMD_DEV_DEL,
1520 		.doit = vdpa_nl_cmd_dev_del_set_doit,
1521 		.flags = GENL_ADMIN_PERM,
1522 	},
1523 	{
1524 		.cmd = VDPA_CMD_DEV_GET,
1525 		.doit = vdpa_nl_cmd_dev_get_doit,
1526 		.dumpit = vdpa_nl_cmd_dev_get_dumpit,
1527 	},
1528 	{
1529 		.cmd = VDPA_CMD_DEV_CONFIG_GET,
1530 		.doit = vdpa_nl_cmd_dev_config_get_doit,
1531 		.dumpit = vdpa_nl_cmd_dev_config_get_dumpit,
1532 	},
1533 	{
1534 		.cmd = VDPA_CMD_DEV_VSTATS_GET,
1535 		.doit = vdpa_nl_cmd_dev_stats_get_doit,
1536 		.flags = GENL_ADMIN_PERM,
1537 	},
1538 	{
1539 		.cmd = VDPA_CMD_DEV_ATTR_SET,
1540 		.doit = vdpa_nl_cmd_dev_attr_set_doit,
1541 		.flags = GENL_ADMIN_PERM,
1542 	},
1543 };
1544 
1545 static struct genl_family vdpa_nl_family __ro_after_init = {
1546 	.name = VDPA_GENL_NAME,
1547 	.version = VDPA_GENL_VERSION,
1548 	.maxattr = VDPA_ATTR_MAX,
1549 	.policy = vdpa_nl_policy,
1550 	.netnsok = false,
1551 	.module = THIS_MODULE,
1552 	.ops = vdpa_nl_ops,
1553 	.n_ops = ARRAY_SIZE(vdpa_nl_ops),
1554 	.resv_start_op = VDPA_CMD_DEV_VSTATS_GET + 1,
1555 };
1556 
vdpa_init(void)1557 static int vdpa_init(void)
1558 {
1559 	int err;
1560 
1561 	err = bus_register(&vdpa_bus);
1562 	if (err)
1563 		return err;
1564 	err = genl_register_family(&vdpa_nl_family);
1565 	if (err)
1566 		goto err;
1567 	return 0;
1568 
1569 err:
1570 	bus_unregister(&vdpa_bus);
1571 	return err;
1572 }
1573 
vdpa_exit(void)1574 static void __exit vdpa_exit(void)
1575 {
1576 	genl_unregister_family(&vdpa_nl_family);
1577 	bus_unregister(&vdpa_bus);
1578 	ida_destroy(&vdpa_index_ida);
1579 }
1580 core_initcall(vdpa_init);
1581 module_exit(vdpa_exit);
1582 
1583 MODULE_AUTHOR("Jason Wang <jasowang@redhat.com>");
1584 MODULE_DESCRIPTION("vDPA bus");
1585 MODULE_LICENSE("GPL v2");
1586