1 /* SPDX-License-Identifier: GPL-2.0-only */
2 /* Copyright (c) 2021-2022, NVIDIA CORPORATION & AFFILIATES
3  */
4 #ifndef __IOMMUFD_PRIVATE_H
5 #define __IOMMUFD_PRIVATE_H
6 
7 #include <linux/rwsem.h>
8 #include <linux/xarray.h>
9 #include <linux/refcount.h>
10 #include <linux/uaccess.h>
11 
12 struct iommu_domain;
13 struct iommu_group;
14 struct iommu_option;
15 struct iommufd_device;
16 
17 struct iommufd_ctx {
18 	struct file *file;
19 	struct xarray objects;
20 	struct xarray groups;
21 
22 	u8 account_mode;
23 	/* Compatibility with VFIO no iommu */
24 	u8 no_iommu_mode;
25 	struct iommufd_ioas *vfio_ioas;
26 };
27 
28 /*
29  * The IOVA to PFN map. The map automatically copies the PFNs into multiple
30  * domains and permits sharing of PFNs between io_pagetable instances. This
31  * supports both a design where IOAS's are 1:1 with a domain (eg because the
32  * domain is HW customized), or where the IOAS is 1:N with multiple generic
33  * domains.  The io_pagetable holds an interval tree of iopt_areas which point
34  * to shared iopt_pages which hold the pfns mapped to the page table.
35  *
36  * The locking order is domains_rwsem -> iova_rwsem -> pages::mutex
37  */
38 struct io_pagetable {
39 	struct rw_semaphore domains_rwsem;
40 	struct xarray domains;
41 	struct xarray access_list;
42 	unsigned int next_domain_id;
43 
44 	struct rw_semaphore iova_rwsem;
45 	struct rb_root_cached area_itree;
46 	/* IOVA that cannot become reserved, struct iopt_allowed */
47 	struct rb_root_cached allowed_itree;
48 	/* IOVA that cannot be allocated, struct iopt_reserved */
49 	struct rb_root_cached reserved_itree;
50 	u8 disable_large_pages;
51 	unsigned long iova_alignment;
52 };
53 
54 void iopt_init_table(struct io_pagetable *iopt);
55 void iopt_destroy_table(struct io_pagetable *iopt);
56 int iopt_get_pages(struct io_pagetable *iopt, unsigned long iova,
57 		   unsigned long length, struct list_head *pages_list);
58 void iopt_free_pages_list(struct list_head *pages_list);
59 enum {
60 	IOPT_ALLOC_IOVA = 1 << 0,
61 };
62 int iopt_map_user_pages(struct iommufd_ctx *ictx, struct io_pagetable *iopt,
63 			unsigned long *iova, void __user *uptr,
64 			unsigned long length, int iommu_prot,
65 			unsigned int flags);
66 int iopt_map_pages(struct io_pagetable *iopt, struct list_head *pages_list,
67 		   unsigned long length, unsigned long *dst_iova,
68 		   int iommu_prot, unsigned int flags);
69 int iopt_unmap_iova(struct io_pagetable *iopt, unsigned long iova,
70 		    unsigned long length, unsigned long *unmapped);
71 int iopt_unmap_all(struct io_pagetable *iopt, unsigned long *unmapped);
72 
73 void iommufd_access_notify_unmap(struct io_pagetable *iopt, unsigned long iova,
74 				 unsigned long length);
75 int iopt_table_add_domain(struct io_pagetable *iopt,
76 			  struct iommu_domain *domain);
77 void iopt_table_remove_domain(struct io_pagetable *iopt,
78 			      struct iommu_domain *domain);
79 int iopt_table_enforce_dev_resv_regions(struct io_pagetable *iopt,
80 					struct device *dev,
81 					phys_addr_t *sw_msi_start);
82 int iopt_set_allow_iova(struct io_pagetable *iopt,
83 			struct rb_root_cached *allowed_iova);
84 int iopt_reserve_iova(struct io_pagetable *iopt, unsigned long start,
85 		      unsigned long last, void *owner);
86 void iopt_remove_reserved_iova(struct io_pagetable *iopt, void *owner);
87 int iopt_cut_iova(struct io_pagetable *iopt, unsigned long *iovas,
88 		  size_t num_iovas);
89 void iopt_enable_large_pages(struct io_pagetable *iopt);
90 int iopt_disable_large_pages(struct io_pagetable *iopt);
91 
92 struct iommufd_ucmd {
93 	struct iommufd_ctx *ictx;
94 	void __user *ubuffer;
95 	u32 user_size;
96 	void *cmd;
97 };
98 
99 int iommufd_vfio_ioctl(struct iommufd_ctx *ictx, unsigned int cmd,
100 		       unsigned long arg);
101 
102 /* Copy the response in ucmd->cmd back to userspace. */
iommufd_ucmd_respond(struct iommufd_ucmd * ucmd,size_t cmd_len)103 static inline int iommufd_ucmd_respond(struct iommufd_ucmd *ucmd,
104 				       size_t cmd_len)
105 {
106 	if (copy_to_user(ucmd->ubuffer, ucmd->cmd,
107 			 min_t(size_t, ucmd->user_size, cmd_len)))
108 		return -EFAULT;
109 	return 0;
110 }
111 
112 enum iommufd_object_type {
113 	IOMMUFD_OBJ_NONE,
114 	IOMMUFD_OBJ_ANY = IOMMUFD_OBJ_NONE,
115 	IOMMUFD_OBJ_DEVICE,
116 	IOMMUFD_OBJ_HW_PAGETABLE,
117 	IOMMUFD_OBJ_IOAS,
118 	IOMMUFD_OBJ_ACCESS,
119 #ifdef CONFIG_IOMMUFD_TEST
120 	IOMMUFD_OBJ_SELFTEST,
121 #endif
122 	IOMMUFD_OBJ_MAX,
123 };
124 
125 /* Base struct for all objects with a userspace ID handle. */
126 struct iommufd_object {
127 	struct rw_semaphore destroy_rwsem;
128 	refcount_t users;
129 	enum iommufd_object_type type;
130 	unsigned int id;
131 };
132 
iommufd_lock_obj(struct iommufd_object * obj)133 static inline bool iommufd_lock_obj(struct iommufd_object *obj)
134 {
135 	if (!down_read_trylock(&obj->destroy_rwsem))
136 		return false;
137 	if (!refcount_inc_not_zero(&obj->users)) {
138 		up_read(&obj->destroy_rwsem);
139 		return false;
140 	}
141 	return true;
142 }
143 
144 struct iommufd_object *iommufd_get_object(struct iommufd_ctx *ictx, u32 id,
145 					  enum iommufd_object_type type);
iommufd_put_object(struct iommufd_object * obj)146 static inline void iommufd_put_object(struct iommufd_object *obj)
147 {
148 	refcount_dec(&obj->users);
149 	up_read(&obj->destroy_rwsem);
150 }
151 
152 void iommufd_object_abort(struct iommufd_ctx *ictx, struct iommufd_object *obj);
153 void iommufd_object_abort_and_destroy(struct iommufd_ctx *ictx,
154 				      struct iommufd_object *obj);
155 void iommufd_object_finalize(struct iommufd_ctx *ictx,
156 			     struct iommufd_object *obj);
157 void __iommufd_object_destroy_user(struct iommufd_ctx *ictx,
158 				   struct iommufd_object *obj, bool allow_fail);
iommufd_object_destroy_user(struct iommufd_ctx * ictx,struct iommufd_object * obj)159 static inline void iommufd_object_destroy_user(struct iommufd_ctx *ictx,
160 					       struct iommufd_object *obj)
161 {
162 	__iommufd_object_destroy_user(ictx, obj, false);
163 }
iommufd_object_deref_user(struct iommufd_ctx * ictx,struct iommufd_object * obj)164 static inline void iommufd_object_deref_user(struct iommufd_ctx *ictx,
165 					     struct iommufd_object *obj)
166 {
167 	__iommufd_object_destroy_user(ictx, obj, true);
168 }
169 
170 struct iommufd_object *_iommufd_object_alloc(struct iommufd_ctx *ictx,
171 					     size_t size,
172 					     enum iommufd_object_type type);
173 
174 #define iommufd_object_alloc(ictx, ptr, type)                                  \
175 	container_of(_iommufd_object_alloc(                                    \
176 			     ictx,                                             \
177 			     sizeof(*(ptr)) + BUILD_BUG_ON_ZERO(               \
178 						      offsetof(typeof(*(ptr)), \
179 							       obj) != 0),     \
180 			     type),                                            \
181 		     typeof(*(ptr)), obj)
182 
183 /*
184  * The IO Address Space (IOAS) pagetable is a virtual page table backed by the
185  * io_pagetable object. It is a user controlled mapping of IOVA -> PFNs. The
186  * mapping is copied into all of the associated domains and made available to
187  * in-kernel users.
188  *
189  * Every iommu_domain that is created is wrapped in a iommufd_hw_pagetable
190  * object. When we go to attach a device to an IOAS we need to get an
191  * iommu_domain and wrapping iommufd_hw_pagetable for it.
192  *
193  * An iommu_domain & iommfd_hw_pagetable will be automatically selected
194  * for a device based on the hwpt_list. If no suitable iommu_domain
195  * is found a new iommu_domain will be created.
196  */
197 struct iommufd_ioas {
198 	struct iommufd_object obj;
199 	struct io_pagetable iopt;
200 	struct mutex mutex;
201 	struct list_head hwpt_list;
202 };
203 
iommufd_get_ioas(struct iommufd_ctx * ictx,u32 id)204 static inline struct iommufd_ioas *iommufd_get_ioas(struct iommufd_ctx *ictx,
205 						    u32 id)
206 {
207 	return container_of(iommufd_get_object(ictx, id,
208 					       IOMMUFD_OBJ_IOAS),
209 			    struct iommufd_ioas, obj);
210 }
211 
212 struct iommufd_ioas *iommufd_ioas_alloc(struct iommufd_ctx *ictx);
213 int iommufd_ioas_alloc_ioctl(struct iommufd_ucmd *ucmd);
214 void iommufd_ioas_destroy(struct iommufd_object *obj);
215 int iommufd_ioas_iova_ranges(struct iommufd_ucmd *ucmd);
216 int iommufd_ioas_allow_iovas(struct iommufd_ucmd *ucmd);
217 int iommufd_ioas_map(struct iommufd_ucmd *ucmd);
218 int iommufd_ioas_copy(struct iommufd_ucmd *ucmd);
219 int iommufd_ioas_unmap(struct iommufd_ucmd *ucmd);
220 int iommufd_ioas_option(struct iommufd_ucmd *ucmd);
221 int iommufd_option_rlimit_mode(struct iommu_option *cmd,
222 			       struct iommufd_ctx *ictx);
223 
224 int iommufd_vfio_ioas(struct iommufd_ucmd *ucmd);
225 
226 /*
227  * A HW pagetable is called an iommu_domain inside the kernel. This user object
228  * allows directly creating and inspecting the domains. Domains that have kernel
229  * owned page tables will be associated with an iommufd_ioas that provides the
230  * IOVA to PFN map.
231  */
232 struct iommufd_hw_pagetable {
233 	struct iommufd_object obj;
234 	struct iommufd_ioas *ioas;
235 	struct iommu_domain *domain;
236 	bool auto_domain : 1;
237 	bool enforce_cache_coherency : 1;
238 	bool msi_cookie : 1;
239 	/* Head at iommufd_ioas::hwpt_list */
240 	struct list_head hwpt_item;
241 };
242 
243 struct iommufd_hw_pagetable *
244 iommufd_hw_pagetable_alloc(struct iommufd_ctx *ictx, struct iommufd_ioas *ioas,
245 			   struct iommufd_device *idev, bool immediate_attach);
246 int iommufd_hw_pagetable_enforce_cc(struct iommufd_hw_pagetable *hwpt);
247 int iommufd_hw_pagetable_attach(struct iommufd_hw_pagetable *hwpt,
248 				struct iommufd_device *idev);
249 struct iommufd_hw_pagetable *
250 iommufd_hw_pagetable_detach(struct iommufd_device *idev);
251 void iommufd_hw_pagetable_destroy(struct iommufd_object *obj);
252 void iommufd_hw_pagetable_abort(struct iommufd_object *obj);
253 int iommufd_hwpt_alloc(struct iommufd_ucmd *ucmd);
254 
iommufd_hw_pagetable_put(struct iommufd_ctx * ictx,struct iommufd_hw_pagetable * hwpt)255 static inline void iommufd_hw_pagetable_put(struct iommufd_ctx *ictx,
256 					    struct iommufd_hw_pagetable *hwpt)
257 {
258 	lockdep_assert_not_held(&hwpt->ioas->mutex);
259 	if (hwpt->auto_domain)
260 		iommufd_object_deref_user(ictx, &hwpt->obj);
261 	else
262 		refcount_dec(&hwpt->obj.users);
263 }
264 
265 struct iommufd_group {
266 	struct kref ref;
267 	struct mutex lock;
268 	struct iommufd_ctx *ictx;
269 	struct iommu_group *group;
270 	struct iommufd_hw_pagetable *hwpt;
271 	struct list_head device_list;
272 	phys_addr_t sw_msi_start;
273 };
274 
275 /*
276  * A iommufd_device object represents the binding relationship between a
277  * consuming driver and the iommufd. These objects are created/destroyed by
278  * external drivers, not by userspace.
279  */
280 struct iommufd_device {
281 	struct iommufd_object obj;
282 	struct iommufd_ctx *ictx;
283 	struct iommufd_group *igroup;
284 	struct list_head group_item;
285 	/* always the physical device */
286 	struct device *dev;
287 	bool enforce_cache_coherency;
288 };
289 
290 static inline struct iommufd_device *
iommufd_get_device(struct iommufd_ucmd * ucmd,u32 id)291 iommufd_get_device(struct iommufd_ucmd *ucmd, u32 id)
292 {
293 	return container_of(iommufd_get_object(ucmd->ictx, id,
294 					       IOMMUFD_OBJ_DEVICE),
295 			    struct iommufd_device, obj);
296 }
297 
298 void iommufd_device_destroy(struct iommufd_object *obj);
299 int iommufd_get_hw_info(struct iommufd_ucmd *ucmd);
300 
301 struct iommufd_access {
302 	struct iommufd_object obj;
303 	struct iommufd_ctx *ictx;
304 	struct iommufd_ioas *ioas;
305 	struct iommufd_ioas *ioas_unpin;
306 	struct mutex ioas_lock;
307 	const struct iommufd_access_ops *ops;
308 	void *data;
309 	unsigned long iova_alignment;
310 	u32 iopt_access_list_id;
311 };
312 
313 int iopt_add_access(struct io_pagetable *iopt, struct iommufd_access *access);
314 void iopt_remove_access(struct io_pagetable *iopt,
315 			struct iommufd_access *access,
316 			u32 iopt_access_list_id);
317 void iommufd_access_destroy_object(struct iommufd_object *obj);
318 
319 #ifdef CONFIG_IOMMUFD_TEST
320 int iommufd_test(struct iommufd_ucmd *ucmd);
321 void iommufd_selftest_destroy(struct iommufd_object *obj);
322 extern size_t iommufd_test_memory_limit;
323 void iommufd_test_syz_conv_iova_id(struct iommufd_ucmd *ucmd,
324 				   unsigned int ioas_id, u64 *iova, u32 *flags);
325 bool iommufd_should_fail(void);
326 int __init iommufd_test_init(void);
327 void iommufd_test_exit(void);
328 bool iommufd_selftest_is_mock_dev(struct device *dev);
329 #else
iommufd_test_syz_conv_iova_id(struct iommufd_ucmd * ucmd,unsigned int ioas_id,u64 * iova,u32 * flags)330 static inline void iommufd_test_syz_conv_iova_id(struct iommufd_ucmd *ucmd,
331 						 unsigned int ioas_id,
332 						 u64 *iova, u32 *flags)
333 {
334 }
iommufd_should_fail(void)335 static inline bool iommufd_should_fail(void)
336 {
337 	return false;
338 }
iommufd_test_init(void)339 static inline int __init iommufd_test_init(void)
340 {
341 	return 0;
342 }
iommufd_test_exit(void)343 static inline void iommufd_test_exit(void)
344 {
345 }
iommufd_selftest_is_mock_dev(struct device * dev)346 static inline bool iommufd_selftest_is_mock_dev(struct device *dev)
347 {
348 	return false;
349 }
350 #endif
351 #endif
352