xref: /linux/drivers/iommu/iommufd/ioas.c (revision a2e33fb92649f4efcc9e81c1d1a1905ca2a76d03)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * Copyright (c) 2021-2022, NVIDIA CORPORATION & AFFILIATES
4  */
5 #include <linux/file.h>
6 #include <linux/interval_tree.h>
7 #include <linux/iommu.h>
8 #include <linux/iommufd.h>
9 #include <uapi/linux/iommufd.h>
10 
11 #include "io_pagetable.h"
12 
iommufd_ioas_destroy(struct iommufd_object * obj)13 void iommufd_ioas_destroy(struct iommufd_object *obj)
14 {
15 	struct iommufd_ioas *ioas = container_of(obj, struct iommufd_ioas, obj);
16 	int rc;
17 
18 	rc = iopt_unmap_all(&ioas->iopt, NULL);
19 	WARN_ON(rc && rc != -ENOENT);
20 	iopt_destroy_table(&ioas->iopt);
21 	mutex_destroy(&ioas->mutex);
22 }
23 
iommufd_ioas_alloc(struct iommufd_ctx * ictx)24 struct iommufd_ioas *iommufd_ioas_alloc(struct iommufd_ctx *ictx)
25 {
26 	struct iommufd_ioas *ioas;
27 
28 	ioas = iommufd_object_alloc(ictx, ioas, IOMMUFD_OBJ_IOAS);
29 	if (IS_ERR(ioas))
30 		return ioas;
31 
32 	iopt_init_table(&ioas->iopt);
33 	INIT_LIST_HEAD(&ioas->hwpt_list);
34 	mutex_init(&ioas->mutex);
35 	return ioas;
36 }
37 
iommufd_ioas_alloc_ioctl(struct iommufd_ucmd * ucmd)38 int iommufd_ioas_alloc_ioctl(struct iommufd_ucmd *ucmd)
39 {
40 	struct iommu_ioas_alloc *cmd = ucmd->cmd;
41 	struct iommufd_ioas *ioas;
42 	int rc;
43 
44 	if (cmd->flags)
45 		return -EOPNOTSUPP;
46 
47 	ioas = iommufd_ioas_alloc(ucmd->ictx);
48 	if (IS_ERR(ioas))
49 		return PTR_ERR(ioas);
50 
51 	cmd->out_ioas_id = ioas->obj.id;
52 	rc = iommufd_ucmd_respond(ucmd, sizeof(*cmd));
53 	if (rc)
54 		goto out_table;
55 
56 	down_read(&ucmd->ictx->ioas_creation_lock);
57 	iommufd_object_finalize(ucmd->ictx, &ioas->obj);
58 	up_read(&ucmd->ictx->ioas_creation_lock);
59 	return 0;
60 
61 out_table:
62 	iommufd_object_abort_and_destroy(ucmd->ictx, &ioas->obj);
63 	return rc;
64 }
65 
iommufd_ioas_iova_ranges(struct iommufd_ucmd * ucmd)66 int iommufd_ioas_iova_ranges(struct iommufd_ucmd *ucmd)
67 {
68 	struct iommu_iova_range __user *ranges;
69 	struct iommu_ioas_iova_ranges *cmd = ucmd->cmd;
70 	struct iommufd_ioas *ioas;
71 	struct interval_tree_span_iter span;
72 	u32 max_iovas;
73 	int rc;
74 
75 	if (cmd->__reserved)
76 		return -EOPNOTSUPP;
77 
78 	ioas = iommufd_get_ioas(ucmd->ictx, cmd->ioas_id);
79 	if (IS_ERR(ioas))
80 		return PTR_ERR(ioas);
81 
82 	down_read(&ioas->iopt.iova_rwsem);
83 	max_iovas = cmd->num_iovas;
84 	ranges = u64_to_user_ptr(cmd->allowed_iovas);
85 	cmd->num_iovas = 0;
86 	cmd->out_iova_alignment = ioas->iopt.iova_alignment;
87 	interval_tree_for_each_span(&span, &ioas->iopt.reserved_itree, 0,
88 				    ULONG_MAX) {
89 		if (!span.is_hole)
90 			continue;
91 		if (cmd->num_iovas < max_iovas) {
92 			struct iommu_iova_range elm = {
93 				.start = span.start_hole,
94 				.last = span.last_hole,
95 			};
96 
97 			if (copy_to_user(&ranges[cmd->num_iovas], &elm,
98 					 sizeof(elm))) {
99 				rc = -EFAULT;
100 				goto out_put;
101 			}
102 		}
103 		cmd->num_iovas++;
104 	}
105 	rc = iommufd_ucmd_respond(ucmd, sizeof(*cmd));
106 	if (rc)
107 		goto out_put;
108 	if (cmd->num_iovas > max_iovas)
109 		rc = -EMSGSIZE;
110 out_put:
111 	up_read(&ioas->iopt.iova_rwsem);
112 	iommufd_put_object(ucmd->ictx, &ioas->obj);
113 	return rc;
114 }
115 
iommufd_ioas_load_iovas(struct rb_root_cached * itree,struct iommu_iova_range __user * ranges,u32 num)116 static int iommufd_ioas_load_iovas(struct rb_root_cached *itree,
117 				   struct iommu_iova_range __user *ranges,
118 				   u32 num)
119 {
120 	u32 i;
121 
122 	for (i = 0; i != num; i++) {
123 		struct iommu_iova_range range;
124 		struct iopt_allowed *allowed;
125 
126 		if (copy_from_user(&range, ranges + i, sizeof(range)))
127 			return -EFAULT;
128 
129 		if (range.start >= range.last)
130 			return -EINVAL;
131 
132 		if (interval_tree_iter_first(itree, range.start, range.last))
133 			return -EINVAL;
134 
135 		allowed = kzalloc(sizeof(*allowed), GFP_KERNEL_ACCOUNT);
136 		if (!allowed)
137 			return -ENOMEM;
138 		allowed->node.start = range.start;
139 		allowed->node.last = range.last;
140 
141 		interval_tree_insert(&allowed->node, itree);
142 	}
143 	return 0;
144 }
145 
iommufd_ioas_allow_iovas(struct iommufd_ucmd * ucmd)146 int iommufd_ioas_allow_iovas(struct iommufd_ucmd *ucmd)
147 {
148 	struct iommu_ioas_allow_iovas *cmd = ucmd->cmd;
149 	struct rb_root_cached allowed_iova = RB_ROOT_CACHED;
150 	struct interval_tree_node *node;
151 	struct iommufd_ioas *ioas;
152 	struct io_pagetable *iopt;
153 	int rc = 0;
154 
155 	if (cmd->__reserved)
156 		return -EOPNOTSUPP;
157 
158 	ioas = iommufd_get_ioas(ucmd->ictx, cmd->ioas_id);
159 	if (IS_ERR(ioas))
160 		return PTR_ERR(ioas);
161 	iopt = &ioas->iopt;
162 
163 	rc = iommufd_ioas_load_iovas(&allowed_iova,
164 				     u64_to_user_ptr(cmd->allowed_iovas),
165 				     cmd->num_iovas);
166 	if (rc)
167 		goto out_free;
168 
169 	/*
170 	 * We want the allowed tree update to be atomic, so we have to keep the
171 	 * original nodes around, and keep track of the new nodes as we allocate
172 	 * memory for them. The simplest solution is to have a new/old tree and
173 	 * then swap new for old. On success we free the old tree, on failure we
174 	 * free the new tree.
175 	 */
176 	rc = iopt_set_allow_iova(iopt, &allowed_iova);
177 out_free:
178 	while ((node = interval_tree_iter_first(&allowed_iova, 0, ULONG_MAX))) {
179 		interval_tree_remove(node, &allowed_iova);
180 		kfree(container_of(node, struct iopt_allowed, node));
181 	}
182 	iommufd_put_object(ucmd->ictx, &ioas->obj);
183 	return rc;
184 }
185 
conv_iommu_prot(u32 map_flags)186 static int conv_iommu_prot(u32 map_flags)
187 {
188 	/*
189 	 * We provide no manual cache coherency ioctls to userspace and most
190 	 * architectures make the CPU ops for cache flushing privileged.
191 	 * Therefore we require the underlying IOMMU to support CPU coherent
192 	 * operation. Support for IOMMU_CACHE is enforced by the
193 	 * IOMMU_CAP_CACHE_COHERENCY test during bind.
194 	 */
195 	int iommu_prot = IOMMU_CACHE;
196 
197 	if (map_flags & IOMMU_IOAS_MAP_WRITEABLE)
198 		iommu_prot |= IOMMU_WRITE;
199 	if (map_flags & IOMMU_IOAS_MAP_READABLE)
200 		iommu_prot |= IOMMU_READ;
201 	return iommu_prot;
202 }
203 
iommufd_ioas_map_file(struct iommufd_ucmd * ucmd)204 int iommufd_ioas_map_file(struct iommufd_ucmd *ucmd)
205 {
206 	struct iommu_ioas_map_file *cmd = ucmd->cmd;
207 	unsigned long iova = cmd->iova;
208 	struct iommufd_ioas *ioas;
209 	unsigned int flags = 0;
210 	struct file *file;
211 	int rc;
212 
213 	if (cmd->flags &
214 	     ~(IOMMU_IOAS_MAP_FIXED_IOVA | IOMMU_IOAS_MAP_WRITEABLE |
215 	       IOMMU_IOAS_MAP_READABLE))
216 		return -EOPNOTSUPP;
217 
218 	if (cmd->iova >= ULONG_MAX || cmd->length >= ULONG_MAX)
219 		return -EOVERFLOW;
220 
221 	if (!(cmd->flags &
222 	      (IOMMU_IOAS_MAP_WRITEABLE | IOMMU_IOAS_MAP_READABLE)))
223 		return -EINVAL;
224 
225 	ioas = iommufd_get_ioas(ucmd->ictx, cmd->ioas_id);
226 	if (IS_ERR(ioas))
227 		return PTR_ERR(ioas);
228 
229 	if (!(cmd->flags & IOMMU_IOAS_MAP_FIXED_IOVA))
230 		flags = IOPT_ALLOC_IOVA;
231 
232 	file = fget(cmd->fd);
233 	if (!file)
234 		return -EBADF;
235 
236 	rc = iopt_map_file_pages(ucmd->ictx, &ioas->iopt, &iova, file,
237 				 cmd->start, cmd->length,
238 				 conv_iommu_prot(cmd->flags), flags);
239 	if (rc)
240 		goto out_put;
241 
242 	cmd->iova = iova;
243 	rc = iommufd_ucmd_respond(ucmd, sizeof(*cmd));
244 out_put:
245 	iommufd_put_object(ucmd->ictx, &ioas->obj);
246 	fput(file);
247 	return rc;
248 }
249 
iommufd_ioas_map(struct iommufd_ucmd * ucmd)250 int iommufd_ioas_map(struct iommufd_ucmd *ucmd)
251 {
252 	struct iommu_ioas_map *cmd = ucmd->cmd;
253 	unsigned long iova = cmd->iova;
254 	struct iommufd_ioas *ioas;
255 	unsigned int flags = 0;
256 	int rc;
257 
258 	if ((cmd->flags &
259 	     ~(IOMMU_IOAS_MAP_FIXED_IOVA | IOMMU_IOAS_MAP_WRITEABLE |
260 	       IOMMU_IOAS_MAP_READABLE)) ||
261 	    cmd->__reserved)
262 		return -EOPNOTSUPP;
263 	if (cmd->iova >= ULONG_MAX || cmd->length >= ULONG_MAX)
264 		return -EOVERFLOW;
265 
266 	if (!(cmd->flags &
267 	      (IOMMU_IOAS_MAP_WRITEABLE | IOMMU_IOAS_MAP_READABLE)))
268 		return -EINVAL;
269 
270 	ioas = iommufd_get_ioas(ucmd->ictx, cmd->ioas_id);
271 	if (IS_ERR(ioas))
272 		return PTR_ERR(ioas);
273 
274 	if (!(cmd->flags & IOMMU_IOAS_MAP_FIXED_IOVA))
275 		flags = IOPT_ALLOC_IOVA;
276 	rc = iopt_map_user_pages(ucmd->ictx, &ioas->iopt, &iova,
277 				 u64_to_user_ptr(cmd->user_va), cmd->length,
278 				 conv_iommu_prot(cmd->flags), flags);
279 	if (rc)
280 		goto out_put;
281 
282 	cmd->iova = iova;
283 	rc = iommufd_ucmd_respond(ucmd, sizeof(*cmd));
284 out_put:
285 	iommufd_put_object(ucmd->ictx, &ioas->obj);
286 	return rc;
287 }
288 
iommufd_ioas_copy(struct iommufd_ucmd * ucmd)289 int iommufd_ioas_copy(struct iommufd_ucmd *ucmd)
290 {
291 	struct iommu_ioas_copy *cmd = ucmd->cmd;
292 	struct iommufd_ioas *src_ioas;
293 	struct iommufd_ioas *dst_ioas;
294 	unsigned int flags = 0;
295 	LIST_HEAD(pages_list);
296 	unsigned long iova;
297 	int rc;
298 
299 	iommufd_test_syz_conv_iova_id(ucmd, cmd->src_ioas_id, &cmd->src_iova,
300 				      &cmd->flags);
301 
302 	if ((cmd->flags &
303 	     ~(IOMMU_IOAS_MAP_FIXED_IOVA | IOMMU_IOAS_MAP_WRITEABLE |
304 	       IOMMU_IOAS_MAP_READABLE)))
305 		return -EOPNOTSUPP;
306 	if (cmd->length >= ULONG_MAX || cmd->src_iova >= ULONG_MAX ||
307 	    cmd->dst_iova >= ULONG_MAX)
308 		return -EOVERFLOW;
309 
310 	if (!(cmd->flags &
311 	      (IOMMU_IOAS_MAP_WRITEABLE | IOMMU_IOAS_MAP_READABLE)))
312 		return -EINVAL;
313 
314 	src_ioas = iommufd_get_ioas(ucmd->ictx, cmd->src_ioas_id);
315 	if (IS_ERR(src_ioas))
316 		return PTR_ERR(src_ioas);
317 	rc = iopt_get_pages(&src_ioas->iopt, cmd->src_iova, cmd->length,
318 			    &pages_list);
319 	iommufd_put_object(ucmd->ictx, &src_ioas->obj);
320 	if (rc)
321 		return rc;
322 
323 	dst_ioas = iommufd_get_ioas(ucmd->ictx, cmd->dst_ioas_id);
324 	if (IS_ERR(dst_ioas)) {
325 		rc = PTR_ERR(dst_ioas);
326 		goto out_pages;
327 	}
328 
329 	if (!(cmd->flags & IOMMU_IOAS_MAP_FIXED_IOVA))
330 		flags = IOPT_ALLOC_IOVA;
331 	iova = cmd->dst_iova;
332 	rc = iopt_map_pages(&dst_ioas->iopt, &pages_list, cmd->length, &iova,
333 			    conv_iommu_prot(cmd->flags), flags);
334 	if (rc)
335 		goto out_put_dst;
336 
337 	cmd->dst_iova = iova;
338 	rc = iommufd_ucmd_respond(ucmd, sizeof(*cmd));
339 out_put_dst:
340 	iommufd_put_object(ucmd->ictx, &dst_ioas->obj);
341 out_pages:
342 	iopt_free_pages_list(&pages_list);
343 	return rc;
344 }
345 
iommufd_ioas_unmap(struct iommufd_ucmd * ucmd)346 int iommufd_ioas_unmap(struct iommufd_ucmd *ucmd)
347 {
348 	struct iommu_ioas_unmap *cmd = ucmd->cmd;
349 	struct iommufd_ioas *ioas;
350 	unsigned long unmapped = 0;
351 	int rc;
352 
353 	ioas = iommufd_get_ioas(ucmd->ictx, cmd->ioas_id);
354 	if (IS_ERR(ioas))
355 		return PTR_ERR(ioas);
356 
357 	if (cmd->iova == 0 && cmd->length == U64_MAX) {
358 		rc = iopt_unmap_all(&ioas->iopt, &unmapped);
359 		if (rc)
360 			goto out_put;
361 	} else {
362 		if (cmd->iova >= ULONG_MAX || cmd->length >= ULONG_MAX) {
363 			rc = -EOVERFLOW;
364 			goto out_put;
365 		}
366 		rc = iopt_unmap_iova(&ioas->iopt, cmd->iova, cmd->length,
367 				     &unmapped);
368 		if (rc)
369 			goto out_put;
370 		if (!unmapped) {
371 			rc = -ENOENT;
372 			goto out_put;
373 		}
374 	}
375 
376 	cmd->length = unmapped;
377 	rc = iommufd_ucmd_respond(ucmd, sizeof(*cmd));
378 
379 out_put:
380 	iommufd_put_object(ucmd->ictx, &ioas->obj);
381 	return rc;
382 }
383 
iommufd_release_all_iova_rwsem(struct iommufd_ctx * ictx,struct xarray * ioas_list)384 static void iommufd_release_all_iova_rwsem(struct iommufd_ctx *ictx,
385 					   struct xarray *ioas_list)
386 {
387 	struct iommufd_ioas *ioas;
388 	unsigned long index;
389 
390 	xa_for_each(ioas_list, index, ioas) {
391 		up_write(&ioas->iopt.iova_rwsem);
392 		refcount_dec(&ioas->obj.users);
393 	}
394 	up_write(&ictx->ioas_creation_lock);
395 	xa_destroy(ioas_list);
396 }
397 
iommufd_take_all_iova_rwsem(struct iommufd_ctx * ictx,struct xarray * ioas_list)398 static int iommufd_take_all_iova_rwsem(struct iommufd_ctx *ictx,
399 				       struct xarray *ioas_list)
400 {
401 	struct iommufd_object *obj;
402 	unsigned long index;
403 	int rc;
404 
405 	/*
406 	 * This is very ugly, it is done instead of adding a lock around
407 	 * pages->source_mm, which is a performance path for mdev, we just
408 	 * obtain the write side of all the iova_rwsems which also protects the
409 	 * pages->source_*. Due to copies we can't know which IOAS could read
410 	 * from the pages, so we just lock everything. This is the only place
411 	 * locks are nested and they are uniformly taken in ID order.
412 	 *
413 	 * ioas_creation_lock prevents new IOAS from being installed in the
414 	 * xarray while we do this, and also prevents more than one thread from
415 	 * holding nested locks.
416 	 */
417 	down_write(&ictx->ioas_creation_lock);
418 	xa_lock(&ictx->objects);
419 	xa_for_each(&ictx->objects, index, obj) {
420 		struct iommufd_ioas *ioas;
421 
422 		if (!obj || obj->type != IOMMUFD_OBJ_IOAS)
423 			continue;
424 
425 		if (!refcount_inc_not_zero(&obj->users))
426 			continue;
427 
428 		xa_unlock(&ictx->objects);
429 
430 		ioas = container_of(obj, struct iommufd_ioas, obj);
431 		down_write_nest_lock(&ioas->iopt.iova_rwsem,
432 				     &ictx->ioas_creation_lock);
433 
434 		rc = xa_err(xa_store(ioas_list, index, ioas, GFP_KERNEL));
435 		if (rc) {
436 			iommufd_release_all_iova_rwsem(ictx, ioas_list);
437 			return rc;
438 		}
439 
440 		xa_lock(&ictx->objects);
441 	}
442 	xa_unlock(&ictx->objects);
443 	return 0;
444 }
445 
need_charge_update(struct iopt_pages * pages)446 static bool need_charge_update(struct iopt_pages *pages)
447 {
448 	switch (pages->account_mode) {
449 	case IOPT_PAGES_ACCOUNT_NONE:
450 		return false;
451 	case IOPT_PAGES_ACCOUNT_MM:
452 		return pages->source_mm != current->mm;
453 	case IOPT_PAGES_ACCOUNT_USER:
454 		/*
455 		 * Update when mm changes because it also accounts
456 		 * in mm->pinned_vm.
457 		 */
458 		return (pages->source_user != current_user()) ||
459 		       (pages->source_mm != current->mm);
460 	}
461 	return true;
462 }
463 
charge_current(unsigned long * npinned)464 static int charge_current(unsigned long *npinned)
465 {
466 	struct iopt_pages tmp = {
467 		.source_mm = current->mm,
468 		.source_task = current->group_leader,
469 		.source_user = current_user(),
470 	};
471 	unsigned int account_mode;
472 	int rc;
473 
474 	for (account_mode = 0; account_mode != IOPT_PAGES_ACCOUNT_MODE_NUM;
475 	     account_mode++) {
476 		if (!npinned[account_mode])
477 			continue;
478 
479 		tmp.account_mode = account_mode;
480 		rc = iopt_pages_update_pinned(&tmp, npinned[account_mode], true,
481 					      NULL);
482 		if (rc)
483 			goto err_undo;
484 	}
485 	return 0;
486 
487 err_undo:
488 	while (account_mode != 0) {
489 		account_mode--;
490 		if (!npinned[account_mode])
491 			continue;
492 		tmp.account_mode = account_mode;
493 		iopt_pages_update_pinned(&tmp, npinned[account_mode], false,
494 					 NULL);
495 	}
496 	return rc;
497 }
498 
change_mm(struct iopt_pages * pages)499 static void change_mm(struct iopt_pages *pages)
500 {
501 	struct task_struct *old_task = pages->source_task;
502 	struct user_struct *old_user = pages->source_user;
503 	struct mm_struct *old_mm = pages->source_mm;
504 
505 	pages->source_mm = current->mm;
506 	mmgrab(pages->source_mm);
507 	mmdrop(old_mm);
508 
509 	pages->source_task = current->group_leader;
510 	get_task_struct(pages->source_task);
511 	put_task_struct(old_task);
512 
513 	pages->source_user = get_uid(current_user());
514 	free_uid(old_user);
515 }
516 
517 #define for_each_ioas_area(_xa, _index, _ioas, _area) \
518 	xa_for_each((_xa), (_index), (_ioas)) \
519 		for (_area = iopt_area_iter_first(&_ioas->iopt, 0, ULONG_MAX); \
520 		     _area; \
521 		     _area = iopt_area_iter_next(_area, 0, ULONG_MAX))
522 
iommufd_ioas_change_process(struct iommufd_ucmd * ucmd)523 int iommufd_ioas_change_process(struct iommufd_ucmd *ucmd)
524 {
525 	struct iommu_ioas_change_process *cmd = ucmd->cmd;
526 	struct iommufd_ctx *ictx = ucmd->ictx;
527 	unsigned long all_npinned[IOPT_PAGES_ACCOUNT_MODE_NUM] = {};
528 	struct iommufd_ioas *ioas;
529 	struct iopt_area *area;
530 	struct iopt_pages *pages;
531 	struct xarray ioas_list;
532 	unsigned long index;
533 	int rc;
534 
535 	if (cmd->__reserved)
536 		return -EOPNOTSUPP;
537 
538 	xa_init(&ioas_list);
539 	rc = iommufd_take_all_iova_rwsem(ictx, &ioas_list);
540 	if (rc)
541 		return rc;
542 
543 	for_each_ioas_area(&ioas_list, index, ioas, area)  {
544 		if (area->pages->type != IOPT_ADDRESS_FILE) {
545 			rc = -EINVAL;
546 			goto out;
547 		}
548 	}
549 
550 	/*
551 	 * Count last_pinned pages, then clear it to avoid double counting
552 	 * if the same iopt_pages is visited multiple times in this loop.
553 	 * Since we are under all the locks, npinned == last_npinned, so we
554 	 * can easily restore last_npinned before we return.
555 	 */
556 	for_each_ioas_area(&ioas_list, index, ioas, area)  {
557 		pages = area->pages;
558 
559 		if (need_charge_update(pages)) {
560 			all_npinned[pages->account_mode] += pages->last_npinned;
561 			pages->last_npinned = 0;
562 		}
563 	}
564 
565 	rc = charge_current(all_npinned);
566 
567 	if (rc) {
568 		/* Charge failed.  Fix last_npinned and bail. */
569 		for_each_ioas_area(&ioas_list, index, ioas, area)
570 			area->pages->last_npinned = area->pages->npinned;
571 		goto out;
572 	}
573 
574 	for_each_ioas_area(&ioas_list, index, ioas, area) {
575 		pages = area->pages;
576 
577 		/* Uncharge the old one (which also restores last_npinned) */
578 		if (need_charge_update(pages)) {
579 			int r = iopt_pages_update_pinned(pages, pages->npinned,
580 							 false, NULL);
581 
582 			if (WARN_ON(r))
583 				rc = r;
584 		}
585 		change_mm(pages);
586 	}
587 
588 out:
589 	iommufd_release_all_iova_rwsem(ictx, &ioas_list);
590 	return rc;
591 }
592 
iommufd_option_rlimit_mode(struct iommu_option * cmd,struct iommufd_ctx * ictx)593 int iommufd_option_rlimit_mode(struct iommu_option *cmd,
594 			       struct iommufd_ctx *ictx)
595 {
596 	if (cmd->object_id)
597 		return -EOPNOTSUPP;
598 
599 	if (cmd->op == IOMMU_OPTION_OP_GET) {
600 		cmd->val64 = ictx->account_mode == IOPT_PAGES_ACCOUNT_MM;
601 		return 0;
602 	}
603 	if (cmd->op == IOMMU_OPTION_OP_SET) {
604 		int rc = 0;
605 
606 		if (!capable(CAP_SYS_RESOURCE))
607 			return -EPERM;
608 
609 		xa_lock(&ictx->objects);
610 		if (!xa_empty(&ictx->objects)) {
611 			rc = -EBUSY;
612 		} else {
613 			if (cmd->val64 == 0)
614 				ictx->account_mode = IOPT_PAGES_ACCOUNT_USER;
615 			else if (cmd->val64 == 1)
616 				ictx->account_mode = IOPT_PAGES_ACCOUNT_MM;
617 			else
618 				rc = -EINVAL;
619 		}
620 		xa_unlock(&ictx->objects);
621 
622 		return rc;
623 	}
624 	return -EOPNOTSUPP;
625 }
626 
iommufd_ioas_option_huge_pages(struct iommu_option * cmd,struct iommufd_ioas * ioas)627 static int iommufd_ioas_option_huge_pages(struct iommu_option *cmd,
628 					  struct iommufd_ioas *ioas)
629 {
630 	if (cmd->op == IOMMU_OPTION_OP_GET) {
631 		cmd->val64 = !ioas->iopt.disable_large_pages;
632 		return 0;
633 	}
634 	if (cmd->op == IOMMU_OPTION_OP_SET) {
635 		if (cmd->val64 == 0)
636 			return iopt_disable_large_pages(&ioas->iopt);
637 		if (cmd->val64 == 1) {
638 			iopt_enable_large_pages(&ioas->iopt);
639 			return 0;
640 		}
641 		return -EINVAL;
642 	}
643 	return -EOPNOTSUPP;
644 }
645 
iommufd_ioas_option(struct iommufd_ucmd * ucmd)646 int iommufd_ioas_option(struct iommufd_ucmd *ucmd)
647 {
648 	struct iommu_option *cmd = ucmd->cmd;
649 	struct iommufd_ioas *ioas;
650 	int rc = 0;
651 
652 	if (cmd->__reserved)
653 		return -EOPNOTSUPP;
654 
655 	ioas = iommufd_get_ioas(ucmd->ictx, cmd->object_id);
656 	if (IS_ERR(ioas))
657 		return PTR_ERR(ioas);
658 
659 	switch (cmd->option_id) {
660 	case IOMMU_OPTION_HUGE_PAGES:
661 		rc = iommufd_ioas_option_huge_pages(cmd, ioas);
662 		break;
663 	default:
664 		rc = -EOPNOTSUPP;
665 	}
666 
667 	iommufd_put_object(ucmd->ictx, &ioas->obj);
668 	return rc;
669 }
670