xref: /openbmc/linux/arch/x86/kernel/cpu/sgx/ioctl.c (revision 54d02069)
1 // SPDX-License-Identifier: GPL-2.0
2 /*  Copyright(c) 2016-20 Intel Corporation. */
3 
4 #include <asm/mman.h>
5 #include <asm/sgx.h>
6 #include <linux/mman.h>
7 #include <linux/delay.h>
8 #include <linux/file.h>
9 #include <linux/hashtable.h>
10 #include <linux/highmem.h>
11 #include <linux/ratelimit.h>
12 #include <linux/sched/signal.h>
13 #include <linux/shmem_fs.h>
14 #include <linux/slab.h>
15 #include <linux/suspend.h>
16 #include "driver.h"
17 #include "encl.h"
18 #include "encls.h"
19 
sgx_encl_grow(struct sgx_encl * encl,bool reclaim)20 struct sgx_va_page *sgx_encl_grow(struct sgx_encl *encl, bool reclaim)
21 {
22 	struct sgx_va_page *va_page = NULL;
23 	void *err;
24 
25 	BUILD_BUG_ON(SGX_VA_SLOT_COUNT !=
26 		(SGX_ENCL_PAGE_VA_OFFSET_MASK >> 3) + 1);
27 
28 	if (!(encl->page_cnt % SGX_VA_SLOT_COUNT)) {
29 		va_page = kzalloc(sizeof(*va_page), GFP_KERNEL);
30 		if (!va_page)
31 			return ERR_PTR(-ENOMEM);
32 
33 		va_page->epc_page = sgx_alloc_va_page(reclaim);
34 		if (IS_ERR(va_page->epc_page)) {
35 			err = ERR_CAST(va_page->epc_page);
36 			kfree(va_page);
37 			return err;
38 		}
39 
40 		WARN_ON_ONCE(encl->page_cnt % SGX_VA_SLOT_COUNT);
41 	}
42 	encl->page_cnt++;
43 	return va_page;
44 }
45 
sgx_encl_shrink(struct sgx_encl * encl,struct sgx_va_page * va_page)46 void sgx_encl_shrink(struct sgx_encl *encl, struct sgx_va_page *va_page)
47 {
48 	encl->page_cnt--;
49 
50 	if (va_page) {
51 		sgx_encl_free_epc_page(va_page->epc_page);
52 		list_del(&va_page->list);
53 		kfree(va_page);
54 	}
55 }
56 
sgx_encl_create(struct sgx_encl * encl,struct sgx_secs * secs)57 static int sgx_encl_create(struct sgx_encl *encl, struct sgx_secs *secs)
58 {
59 	struct sgx_epc_page *secs_epc;
60 	struct sgx_va_page *va_page;
61 	struct sgx_pageinfo pginfo;
62 	struct sgx_secinfo secinfo;
63 	unsigned long encl_size;
64 	struct file *backing;
65 	long ret;
66 
67 	va_page = sgx_encl_grow(encl, true);
68 	if (IS_ERR(va_page))
69 		return PTR_ERR(va_page);
70 	else if (va_page)
71 		list_add(&va_page->list, &encl->va_pages);
72 	/* else the tail page of the VA page list had free slots. */
73 
74 	/* The extra page goes to SECS. */
75 	encl_size = secs->size + PAGE_SIZE;
76 
77 	backing = shmem_file_setup("SGX backing", encl_size + (encl_size >> 5),
78 				   VM_NORESERVE);
79 	if (IS_ERR(backing)) {
80 		ret = PTR_ERR(backing);
81 		goto err_out_shrink;
82 	}
83 
84 	encl->backing = backing;
85 
86 	secs_epc = sgx_alloc_epc_page(&encl->secs, true);
87 	if (IS_ERR(secs_epc)) {
88 		ret = PTR_ERR(secs_epc);
89 		goto err_out_backing;
90 	}
91 
92 	encl->secs.epc_page = secs_epc;
93 
94 	pginfo.addr = 0;
95 	pginfo.contents = (unsigned long)secs;
96 	pginfo.metadata = (unsigned long)&secinfo;
97 	pginfo.secs = 0;
98 	memset(&secinfo, 0, sizeof(secinfo));
99 
100 	ret = __ecreate((void *)&pginfo, sgx_get_epc_virt_addr(secs_epc));
101 	if (ret) {
102 		ret = -EIO;
103 		goto err_out;
104 	}
105 
106 	if (secs->attributes & SGX_ATTR_DEBUG)
107 		set_bit(SGX_ENCL_DEBUG, &encl->flags);
108 
109 	encl->secs.encl = encl;
110 	encl->secs.type = SGX_PAGE_TYPE_SECS;
111 	encl->base = secs->base;
112 	encl->size = secs->size;
113 	encl->attributes = secs->attributes;
114 	encl->attributes_mask = SGX_ATTR_UNPRIV_MASK;
115 
116 	/* Set only after completion, as encl->lock has not been taken. */
117 	set_bit(SGX_ENCL_CREATED, &encl->flags);
118 
119 	return 0;
120 
121 err_out:
122 	sgx_encl_free_epc_page(encl->secs.epc_page);
123 	encl->secs.epc_page = NULL;
124 
125 err_out_backing:
126 	fput(encl->backing);
127 	encl->backing = NULL;
128 
129 err_out_shrink:
130 	sgx_encl_shrink(encl, va_page);
131 
132 	return ret;
133 }
134 
135 /**
136  * sgx_ioc_enclave_create() - handler for %SGX_IOC_ENCLAVE_CREATE
137  * @encl:	An enclave pointer.
138  * @arg:	The ioctl argument.
139  *
140  * Allocate kernel data structures for the enclave and invoke ECREATE.
141  *
142  * Return:
143  * - 0:		Success.
144  * - -EIO:	ECREATE failed.
145  * - -errno:	POSIX error.
146  */
sgx_ioc_enclave_create(struct sgx_encl * encl,void __user * arg)147 static long sgx_ioc_enclave_create(struct sgx_encl *encl, void __user *arg)
148 {
149 	struct sgx_enclave_create create_arg;
150 	void *secs;
151 	int ret;
152 
153 	if (test_bit(SGX_ENCL_CREATED, &encl->flags))
154 		return -EINVAL;
155 
156 	if (copy_from_user(&create_arg, arg, sizeof(create_arg)))
157 		return -EFAULT;
158 
159 	secs = kmalloc(PAGE_SIZE, GFP_KERNEL);
160 	if (!secs)
161 		return -ENOMEM;
162 
163 	if (copy_from_user(secs, (void __user *)create_arg.src, PAGE_SIZE))
164 		ret = -EFAULT;
165 	else
166 		ret = sgx_encl_create(encl, secs);
167 
168 	kfree(secs);
169 	return ret;
170 }
171 
sgx_validate_secinfo(struct sgx_secinfo * secinfo)172 static int sgx_validate_secinfo(struct sgx_secinfo *secinfo)
173 {
174 	u64 perm = secinfo->flags & SGX_SECINFO_PERMISSION_MASK;
175 	u64 pt   = secinfo->flags & SGX_SECINFO_PAGE_TYPE_MASK;
176 
177 	if (pt != SGX_SECINFO_REG && pt != SGX_SECINFO_TCS)
178 		return -EINVAL;
179 
180 	if ((perm & SGX_SECINFO_W) && !(perm & SGX_SECINFO_R))
181 		return -EINVAL;
182 
183 	/*
184 	 * CPU will silently overwrite the permissions as zero, which means
185 	 * that we need to validate it ourselves.
186 	 */
187 	if (pt == SGX_SECINFO_TCS && perm)
188 		return -EINVAL;
189 
190 	if (secinfo->flags & SGX_SECINFO_RESERVED_MASK)
191 		return -EINVAL;
192 
193 	if (memchr_inv(secinfo->reserved, 0, sizeof(secinfo->reserved)))
194 		return -EINVAL;
195 
196 	return 0;
197 }
198 
__sgx_encl_add_page(struct sgx_encl * encl,struct sgx_encl_page * encl_page,struct sgx_epc_page * epc_page,struct sgx_secinfo * secinfo,unsigned long src)199 static int __sgx_encl_add_page(struct sgx_encl *encl,
200 			       struct sgx_encl_page *encl_page,
201 			       struct sgx_epc_page *epc_page,
202 			       struct sgx_secinfo *secinfo, unsigned long src)
203 {
204 	struct sgx_pageinfo pginfo;
205 	struct vm_area_struct *vma;
206 	struct page *src_page;
207 	int ret;
208 
209 	/* Deny noexec. */
210 	vma = find_vma(current->mm, src);
211 	if (!vma)
212 		return -EFAULT;
213 
214 	if (!(vma->vm_flags & VM_MAYEXEC))
215 		return -EACCES;
216 
217 	ret = get_user_pages(src, 1, 0, &src_page);
218 	if (ret < 1)
219 		return -EFAULT;
220 
221 	pginfo.secs = (unsigned long)sgx_get_epc_virt_addr(encl->secs.epc_page);
222 	pginfo.addr = encl_page->desc & PAGE_MASK;
223 	pginfo.metadata = (unsigned long)secinfo;
224 	pginfo.contents = (unsigned long)kmap_local_page(src_page);
225 
226 	ret = __eadd(&pginfo, sgx_get_epc_virt_addr(epc_page));
227 
228 	kunmap_local((void *)pginfo.contents);
229 	put_page(src_page);
230 
231 	return ret ? -EIO : 0;
232 }
233 
234 /*
235  * If the caller requires measurement of the page as a proof for the content,
236  * use EEXTEND to add a measurement for 256 bytes of the page. Repeat this
237  * operation until the entire page is measured."
238  */
__sgx_encl_extend(struct sgx_encl * encl,struct sgx_epc_page * epc_page)239 static int __sgx_encl_extend(struct sgx_encl *encl,
240 			     struct sgx_epc_page *epc_page)
241 {
242 	unsigned long offset;
243 	int ret;
244 
245 	for (offset = 0; offset < PAGE_SIZE; offset += SGX_EEXTEND_BLOCK_SIZE) {
246 		ret = __eextend(sgx_get_epc_virt_addr(encl->secs.epc_page),
247 				sgx_get_epc_virt_addr(epc_page) + offset);
248 		if (ret) {
249 			if (encls_failed(ret))
250 				ENCLS_WARN(ret, "EEXTEND");
251 
252 			return -EIO;
253 		}
254 	}
255 
256 	return 0;
257 }
258 
sgx_encl_add_page(struct sgx_encl * encl,unsigned long src,unsigned long offset,struct sgx_secinfo * secinfo,unsigned long flags)259 static int sgx_encl_add_page(struct sgx_encl *encl, unsigned long src,
260 			     unsigned long offset, struct sgx_secinfo *secinfo,
261 			     unsigned long flags)
262 {
263 	struct sgx_encl_page *encl_page;
264 	struct sgx_epc_page *epc_page;
265 	struct sgx_va_page *va_page;
266 	int ret;
267 
268 	encl_page = sgx_encl_page_alloc(encl, offset, secinfo->flags);
269 	if (IS_ERR(encl_page))
270 		return PTR_ERR(encl_page);
271 
272 	epc_page = sgx_alloc_epc_page(encl_page, true);
273 	if (IS_ERR(epc_page)) {
274 		kfree(encl_page);
275 		return PTR_ERR(epc_page);
276 	}
277 
278 	va_page = sgx_encl_grow(encl, true);
279 	if (IS_ERR(va_page)) {
280 		ret = PTR_ERR(va_page);
281 		goto err_out_free;
282 	}
283 
284 	mmap_read_lock(current->mm);
285 	mutex_lock(&encl->lock);
286 
287 	/*
288 	 * Adding to encl->va_pages must be done under encl->lock.  Ditto for
289 	 * deleting (via sgx_encl_shrink()) in the error path.
290 	 */
291 	if (va_page)
292 		list_add(&va_page->list, &encl->va_pages);
293 
294 	/*
295 	 * Insert prior to EADD in case of OOM.  EADD modifies MRENCLAVE, i.e.
296 	 * can't be gracefully unwound, while failure on EADD/EXTEND is limited
297 	 * to userspace errors (or kernel/hardware bugs).
298 	 */
299 	ret = xa_insert(&encl->page_array, PFN_DOWN(encl_page->desc),
300 			encl_page, GFP_KERNEL);
301 	if (ret)
302 		goto err_out_unlock;
303 
304 	ret = __sgx_encl_add_page(encl, encl_page, epc_page, secinfo,
305 				  src);
306 	if (ret)
307 		goto err_out;
308 
309 	/*
310 	 * Complete the "add" before doing the "extend" so that the "add"
311 	 * isn't in a half-baked state in the extremely unlikely scenario
312 	 * the enclave will be destroyed in response to EEXTEND failure.
313 	 */
314 	encl_page->encl = encl;
315 	encl_page->epc_page = epc_page;
316 	encl_page->type = (secinfo->flags & SGX_SECINFO_PAGE_TYPE_MASK) >> 8;
317 	encl->secs_child_cnt++;
318 
319 	if (flags & SGX_PAGE_MEASURE) {
320 		ret = __sgx_encl_extend(encl, epc_page);
321 		if (ret)
322 			goto err_out;
323 	}
324 
325 	sgx_mark_page_reclaimable(encl_page->epc_page);
326 	mutex_unlock(&encl->lock);
327 	mmap_read_unlock(current->mm);
328 	return ret;
329 
330 err_out:
331 	xa_erase(&encl->page_array, PFN_DOWN(encl_page->desc));
332 
333 err_out_unlock:
334 	sgx_encl_shrink(encl, va_page);
335 	mutex_unlock(&encl->lock);
336 	mmap_read_unlock(current->mm);
337 
338 err_out_free:
339 	sgx_encl_free_epc_page(epc_page);
340 	kfree(encl_page);
341 
342 	return ret;
343 }
344 
345 /*
346  * Ensure user provided offset and length values are valid for
347  * an enclave.
348  */
sgx_validate_offset_length(struct sgx_encl * encl,unsigned long offset,unsigned long length)349 static int sgx_validate_offset_length(struct sgx_encl *encl,
350 				      unsigned long offset,
351 				      unsigned long length)
352 {
353 	if (!IS_ALIGNED(offset, PAGE_SIZE))
354 		return -EINVAL;
355 
356 	if (!length || !IS_ALIGNED(length, PAGE_SIZE))
357 		return -EINVAL;
358 
359 	if (offset + length < offset)
360 		return -EINVAL;
361 
362 	if (offset + length - PAGE_SIZE >= encl->size)
363 		return -EINVAL;
364 
365 	return 0;
366 }
367 
368 /**
369  * sgx_ioc_enclave_add_pages() - The handler for %SGX_IOC_ENCLAVE_ADD_PAGES
370  * @encl:       an enclave pointer
371  * @arg:	a user pointer to a struct sgx_enclave_add_pages instance
372  *
373  * Add one or more pages to an uninitialized enclave, and optionally extend the
374  * measurement with the contents of the page. The SECINFO and measurement mask
375  * are applied to all pages.
376  *
377  * A SECINFO for a TCS is required to always contain zero permissions because
378  * CPU silently zeros them. Allowing anything else would cause a mismatch in
379  * the measurement.
380  *
381  * mmap()'s protection bits are capped by the page permissions. For each page
382  * address, the maximum protection bits are computed with the following
383  * heuristics:
384  *
385  * 1. A regular page: PROT_R, PROT_W and PROT_X match the SECINFO permissions.
386  * 2. A TCS page: PROT_R | PROT_W.
387  *
388  * mmap() is not allowed to surpass the minimum of the maximum protection bits
389  * within the given address range.
390  *
391  * The function deinitializes kernel data structures for enclave and returns
392  * -EIO in any of the following conditions:
393  *
394  * - Enclave Page Cache (EPC), the physical memory holding enclaves, has
395  *   been invalidated. This will cause EADD and EEXTEND to fail.
396  * - If the source address is corrupted somehow when executing EADD.
397  *
398  * Return:
399  * - 0:		Success.
400  * - -EACCES:	The source page is located in a noexec partition.
401  * - -ENOMEM:	Out of EPC pages.
402  * - -EINTR:	The call was interrupted before data was processed.
403  * - -EIO:	Either EADD or EEXTEND failed because invalid source address
404  *		or power cycle.
405  * - -errno:	POSIX error.
406  */
sgx_ioc_enclave_add_pages(struct sgx_encl * encl,void __user * arg)407 static long sgx_ioc_enclave_add_pages(struct sgx_encl *encl, void __user *arg)
408 {
409 	struct sgx_enclave_add_pages add_arg;
410 	struct sgx_secinfo secinfo;
411 	unsigned long c;
412 	int ret;
413 
414 	if (!test_bit(SGX_ENCL_CREATED, &encl->flags) ||
415 	    test_bit(SGX_ENCL_INITIALIZED, &encl->flags))
416 		return -EINVAL;
417 
418 	if (copy_from_user(&add_arg, arg, sizeof(add_arg)))
419 		return -EFAULT;
420 
421 	if (!IS_ALIGNED(add_arg.src, PAGE_SIZE))
422 		return -EINVAL;
423 
424 	if (sgx_validate_offset_length(encl, add_arg.offset, add_arg.length))
425 		return -EINVAL;
426 
427 	if (copy_from_user(&secinfo, (void __user *)add_arg.secinfo,
428 			   sizeof(secinfo)))
429 		return -EFAULT;
430 
431 	if (sgx_validate_secinfo(&secinfo))
432 		return -EINVAL;
433 
434 	for (c = 0 ; c < add_arg.length; c += PAGE_SIZE) {
435 		if (signal_pending(current)) {
436 			if (!c)
437 				ret = -ERESTARTSYS;
438 
439 			break;
440 		}
441 
442 		if (need_resched())
443 			cond_resched();
444 
445 		ret = sgx_encl_add_page(encl, add_arg.src + c, add_arg.offset + c,
446 					&secinfo, add_arg.flags);
447 		if (ret)
448 			break;
449 	}
450 
451 	add_arg.count = c;
452 
453 	if (copy_to_user(arg, &add_arg, sizeof(add_arg)))
454 		return -EFAULT;
455 
456 	return ret;
457 }
458 
__sgx_get_key_hash(struct crypto_shash * tfm,const void * modulus,void * hash)459 static int __sgx_get_key_hash(struct crypto_shash *tfm, const void *modulus,
460 			      void *hash)
461 {
462 	SHASH_DESC_ON_STACK(shash, tfm);
463 
464 	shash->tfm = tfm;
465 
466 	return crypto_shash_digest(shash, modulus, SGX_MODULUS_SIZE, hash);
467 }
468 
sgx_get_key_hash(const void * modulus,void * hash)469 static int sgx_get_key_hash(const void *modulus, void *hash)
470 {
471 	struct crypto_shash *tfm;
472 	int ret;
473 
474 	tfm = crypto_alloc_shash("sha256", 0, CRYPTO_ALG_ASYNC);
475 	if (IS_ERR(tfm))
476 		return PTR_ERR(tfm);
477 
478 	ret = __sgx_get_key_hash(tfm, modulus, hash);
479 
480 	crypto_free_shash(tfm);
481 	return ret;
482 }
483 
sgx_encl_init(struct sgx_encl * encl,struct sgx_sigstruct * sigstruct,void * token)484 static int sgx_encl_init(struct sgx_encl *encl, struct sgx_sigstruct *sigstruct,
485 			 void *token)
486 {
487 	u64 mrsigner[4];
488 	int i, j;
489 	void *addr;
490 	int ret;
491 
492 	/*
493 	 * Deny initializing enclaves with attributes (namely provisioning)
494 	 * that have not been explicitly allowed.
495 	 */
496 	if (encl->attributes & ~encl->attributes_mask)
497 		return -EACCES;
498 
499 	/*
500 	 * Attributes should not be enforced *only* against what's available on
501 	 * platform (done in sgx_encl_create) but checked and enforced against
502 	 * the mask for enforcement in sigstruct. For example an enclave could
503 	 * opt to sign with AVX bit in xfrm, but still be loadable on a platform
504 	 * without it if the sigstruct->body.attributes_mask does not turn that
505 	 * bit on.
506 	 */
507 	if (sigstruct->body.attributes & sigstruct->body.attributes_mask &
508 	    sgx_attributes_reserved_mask)
509 		return -EINVAL;
510 
511 	if (sigstruct->body.miscselect & sigstruct->body.misc_mask &
512 	    sgx_misc_reserved_mask)
513 		return -EINVAL;
514 
515 	if (sigstruct->body.xfrm & sigstruct->body.xfrm_mask &
516 	    sgx_xfrm_reserved_mask)
517 		return -EINVAL;
518 
519 	ret = sgx_get_key_hash(sigstruct->modulus, mrsigner);
520 	if (ret)
521 		return ret;
522 
523 	mutex_lock(&encl->lock);
524 
525 	/*
526 	 * ENCLS[EINIT] is interruptible because it has such a high latency,
527 	 * e.g. 50k+ cycles on success. If an IRQ/NMI/SMI becomes pending,
528 	 * EINIT may fail with SGX_UNMASKED_EVENT so that the event can be
529 	 * serviced.
530 	 */
531 	for (i = 0; i < SGX_EINIT_SLEEP_COUNT; i++) {
532 		for (j = 0; j < SGX_EINIT_SPIN_COUNT; j++) {
533 			addr = sgx_get_epc_virt_addr(encl->secs.epc_page);
534 
535 			preempt_disable();
536 
537 			sgx_update_lepubkeyhash(mrsigner);
538 
539 			ret = __einit(sigstruct, token, addr);
540 
541 			preempt_enable();
542 
543 			if (ret == SGX_UNMASKED_EVENT)
544 				continue;
545 			else
546 				break;
547 		}
548 
549 		if (ret != SGX_UNMASKED_EVENT)
550 			break;
551 
552 		msleep_interruptible(SGX_EINIT_SLEEP_TIME);
553 
554 		if (signal_pending(current)) {
555 			ret = -ERESTARTSYS;
556 			goto err_out;
557 		}
558 	}
559 
560 	if (encls_faulted(ret)) {
561 		if (encls_failed(ret))
562 			ENCLS_WARN(ret, "EINIT");
563 
564 		ret = -EIO;
565 	} else if (ret) {
566 		pr_debug("EINIT returned %d\n", ret);
567 		ret = -EPERM;
568 	} else {
569 		set_bit(SGX_ENCL_INITIALIZED, &encl->flags);
570 	}
571 
572 err_out:
573 	mutex_unlock(&encl->lock);
574 	return ret;
575 }
576 
577 /**
578  * sgx_ioc_enclave_init() - handler for %SGX_IOC_ENCLAVE_INIT
579  * @encl:	an enclave pointer
580  * @arg:	userspace pointer to a struct sgx_enclave_init instance
581  *
582  * Flush any outstanding enqueued EADD operations and perform EINIT.  The
583  * Launch Enclave Public Key Hash MSRs are rewritten as necessary to match
584  * the enclave's MRSIGNER, which is caculated from the provided sigstruct.
585  *
586  * Return:
587  * - 0:		Success.
588  * - -EPERM:	Invalid SIGSTRUCT.
589  * - -EIO:	EINIT failed because of a power cycle.
590  * - -errno:	POSIX error.
591  */
sgx_ioc_enclave_init(struct sgx_encl * encl,void __user * arg)592 static long sgx_ioc_enclave_init(struct sgx_encl *encl, void __user *arg)
593 {
594 	struct sgx_sigstruct *sigstruct;
595 	struct sgx_enclave_init init_arg;
596 	void *token;
597 	int ret;
598 
599 	if (!test_bit(SGX_ENCL_CREATED, &encl->flags) ||
600 	    test_bit(SGX_ENCL_INITIALIZED, &encl->flags))
601 		return -EINVAL;
602 
603 	if (copy_from_user(&init_arg, arg, sizeof(init_arg)))
604 		return -EFAULT;
605 
606 	/*
607 	 * 'sigstruct' must be on a page boundary and 'token' on a 512 byte
608 	 * boundary.  kmalloc() will give this alignment when allocating
609 	 * PAGE_SIZE bytes.
610 	 */
611 	sigstruct = kmalloc(PAGE_SIZE, GFP_KERNEL);
612 	if (!sigstruct)
613 		return -ENOMEM;
614 
615 	token = (void *)((unsigned long)sigstruct + PAGE_SIZE / 2);
616 	memset(token, 0, SGX_LAUNCH_TOKEN_SIZE);
617 
618 	if (copy_from_user(sigstruct, (void __user *)init_arg.sigstruct,
619 			   sizeof(*sigstruct))) {
620 		ret = -EFAULT;
621 		goto out;
622 	}
623 
624 	/*
625 	 * A legacy field used with Intel signed enclaves. These used to mean
626 	 * regular and architectural enclaves. The CPU only accepts these values
627 	 * but they do not have any other meaning.
628 	 *
629 	 * Thus, reject any other values.
630 	 */
631 	if (sigstruct->header.vendor != 0x0000 &&
632 	    sigstruct->header.vendor != 0x8086) {
633 		ret = -EINVAL;
634 		goto out;
635 	}
636 
637 	ret = sgx_encl_init(encl, sigstruct, token);
638 
639 out:
640 	kfree(sigstruct);
641 	return ret;
642 }
643 
644 /**
645  * sgx_ioc_enclave_provision() - handler for %SGX_IOC_ENCLAVE_PROVISION
646  * @encl:	an enclave pointer
647  * @arg:	userspace pointer to a struct sgx_enclave_provision instance
648  *
649  * Allow ATTRIBUTE.PROVISION_KEY for an enclave by providing a file handle to
650  * /dev/sgx_provision.
651  *
652  * Return:
653  * - 0:		Success.
654  * - -errno:	Otherwise.
655  */
sgx_ioc_enclave_provision(struct sgx_encl * encl,void __user * arg)656 static long sgx_ioc_enclave_provision(struct sgx_encl *encl, void __user *arg)
657 {
658 	struct sgx_enclave_provision params;
659 
660 	if (copy_from_user(&params, arg, sizeof(params)))
661 		return -EFAULT;
662 
663 	return sgx_set_attribute(&encl->attributes_mask, params.fd);
664 }
665 
666 /*
667  * Ensure enclave is ready for SGX2 functions. Readiness is checked
668  * by ensuring the hardware supports SGX2 and the enclave is initialized
669  * and thus able to handle requests to modify pages within it.
670  */
sgx_ioc_sgx2_ready(struct sgx_encl * encl)671 static int sgx_ioc_sgx2_ready(struct sgx_encl *encl)
672 {
673 	if (!(cpu_feature_enabled(X86_FEATURE_SGX2)))
674 		return -ENODEV;
675 
676 	if (!test_bit(SGX_ENCL_INITIALIZED, &encl->flags))
677 		return -EINVAL;
678 
679 	return 0;
680 }
681 
682 /*
683  * Some SGX functions require that no cached linear-to-physical address
684  * mappings are present before they can succeed. Collaborate with
685  * hardware via ENCLS[ETRACK] to ensure that all cached
686  * linear-to-physical address mappings belonging to all threads of
687  * the enclave are cleared. See sgx_encl_cpumask() for details.
688  *
689  * Must be called with enclave's mutex held from the time the
690  * SGX function requiring that no cached linear-to-physical mappings
691  * are present is executed until this ETRACK flow is complete.
692  */
sgx_enclave_etrack(struct sgx_encl * encl)693 static int sgx_enclave_etrack(struct sgx_encl *encl)
694 {
695 	void *epc_virt;
696 	int ret;
697 
698 	epc_virt = sgx_get_epc_virt_addr(encl->secs.epc_page);
699 	ret = __etrack(epc_virt);
700 	if (ret) {
701 		/*
702 		 * ETRACK only fails when there is an OS issue. For
703 		 * example, two consecutive ETRACK was sent without
704 		 * completed IPI between.
705 		 */
706 		pr_err_once("ETRACK returned %d (0x%x)", ret, ret);
707 		/*
708 		 * Send IPIs to kick CPUs out of the enclave and
709 		 * try ETRACK again.
710 		 */
711 		on_each_cpu_mask(sgx_encl_cpumask(encl), sgx_ipi_cb, NULL, 1);
712 		ret = __etrack(epc_virt);
713 		if (ret) {
714 			pr_err_once("ETRACK repeat returned %d (0x%x)",
715 				    ret, ret);
716 			return -EFAULT;
717 		}
718 	}
719 	on_each_cpu_mask(sgx_encl_cpumask(encl), sgx_ipi_cb, NULL, 1);
720 
721 	return 0;
722 }
723 
724 /**
725  * sgx_enclave_restrict_permissions() - Restrict EPCM permissions
726  * @encl:	Enclave to which the pages belong.
727  * @modp:	Checked parameters from user on which pages need modifying and
728  *              their new permissions.
729  *
730  * Return:
731  * - 0:		Success.
732  * - -errno:	Otherwise.
733  */
734 static long
sgx_enclave_restrict_permissions(struct sgx_encl * encl,struct sgx_enclave_restrict_permissions * modp)735 sgx_enclave_restrict_permissions(struct sgx_encl *encl,
736 				 struct sgx_enclave_restrict_permissions *modp)
737 {
738 	struct sgx_encl_page *entry;
739 	struct sgx_secinfo secinfo;
740 	unsigned long addr;
741 	unsigned long c;
742 	void *epc_virt;
743 	int ret;
744 
745 	memset(&secinfo, 0, sizeof(secinfo));
746 	secinfo.flags = modp->permissions & SGX_SECINFO_PERMISSION_MASK;
747 
748 	for (c = 0 ; c < modp->length; c += PAGE_SIZE) {
749 		addr = encl->base + modp->offset + c;
750 
751 		sgx_reclaim_direct();
752 
753 		mutex_lock(&encl->lock);
754 
755 		entry = sgx_encl_load_page(encl, addr);
756 		if (IS_ERR(entry)) {
757 			ret = PTR_ERR(entry) == -EBUSY ? -EAGAIN : -EFAULT;
758 			goto out_unlock;
759 		}
760 
761 		/*
762 		 * Changing EPCM permissions is only supported on regular
763 		 * SGX pages. Attempting this change on other pages will
764 		 * result in #PF.
765 		 */
766 		if (entry->type != SGX_PAGE_TYPE_REG) {
767 			ret = -EINVAL;
768 			goto out_unlock;
769 		}
770 
771 		/*
772 		 * Apart from ensuring that read-access remains, do not verify
773 		 * the permission bits requested. Kernel has no control over
774 		 * how EPCM permissions can be relaxed from within the enclave.
775 		 * ENCLS[EMODPR] can only remove existing EPCM permissions,
776 		 * attempting to set new permissions will be ignored by the
777 		 * hardware.
778 		 */
779 
780 		/* Change EPCM permissions. */
781 		epc_virt = sgx_get_epc_virt_addr(entry->epc_page);
782 		ret = __emodpr(&secinfo, epc_virt);
783 		if (encls_faulted(ret)) {
784 			/*
785 			 * All possible faults should be avoidable:
786 			 * parameters have been checked, will only change
787 			 * permissions of a regular page, and no concurrent
788 			 * SGX1/SGX2 ENCLS instructions since these
789 			 * are protected with mutex.
790 			 */
791 			pr_err_once("EMODPR encountered exception %d\n",
792 				    ENCLS_TRAPNR(ret));
793 			ret = -EFAULT;
794 			goto out_unlock;
795 		}
796 		if (encls_failed(ret)) {
797 			modp->result = ret;
798 			ret = -EFAULT;
799 			goto out_unlock;
800 		}
801 
802 		ret = sgx_enclave_etrack(encl);
803 		if (ret) {
804 			ret = -EFAULT;
805 			goto out_unlock;
806 		}
807 
808 		mutex_unlock(&encl->lock);
809 	}
810 
811 	ret = 0;
812 	goto out;
813 
814 out_unlock:
815 	mutex_unlock(&encl->lock);
816 out:
817 	modp->count = c;
818 
819 	return ret;
820 }
821 
822 /**
823  * sgx_ioc_enclave_restrict_permissions() - handler for
824  *                                        %SGX_IOC_ENCLAVE_RESTRICT_PERMISSIONS
825  * @encl:	an enclave pointer
826  * @arg:	userspace pointer to a &struct sgx_enclave_restrict_permissions
827  *		instance
828  *
829  * SGX2 distinguishes between relaxing and restricting the enclave page
830  * permissions maintained by the hardware (EPCM permissions) of pages
831  * belonging to an initialized enclave (after SGX_IOC_ENCLAVE_INIT).
832  *
833  * EPCM permissions cannot be restricted from within the enclave, the enclave
834  * requires the kernel to run the privileged level 0 instructions ENCLS[EMODPR]
835  * and ENCLS[ETRACK]. An attempt to relax EPCM permissions with this call
836  * will be ignored by the hardware.
837  *
838  * Return:
839  * - 0:		Success
840  * - -errno:	Otherwise
841  */
sgx_ioc_enclave_restrict_permissions(struct sgx_encl * encl,void __user * arg)842 static long sgx_ioc_enclave_restrict_permissions(struct sgx_encl *encl,
843 						 void __user *arg)
844 {
845 	struct sgx_enclave_restrict_permissions params;
846 	long ret;
847 
848 	ret = sgx_ioc_sgx2_ready(encl);
849 	if (ret)
850 		return ret;
851 
852 	if (copy_from_user(&params, arg, sizeof(params)))
853 		return -EFAULT;
854 
855 	if (sgx_validate_offset_length(encl, params.offset, params.length))
856 		return -EINVAL;
857 
858 	if (params.permissions & ~SGX_SECINFO_PERMISSION_MASK)
859 		return -EINVAL;
860 
861 	/*
862 	 * Fail early if invalid permissions requested to prevent ENCLS[EMODPR]
863 	 * from faulting later when the CPU does the same check.
864 	 */
865 	if ((params.permissions & SGX_SECINFO_W) &&
866 	    !(params.permissions & SGX_SECINFO_R))
867 		return -EINVAL;
868 
869 	if (params.result || params.count)
870 		return -EINVAL;
871 
872 	ret = sgx_enclave_restrict_permissions(encl, &params);
873 
874 	if (copy_to_user(arg, &params, sizeof(params)))
875 		return -EFAULT;
876 
877 	return ret;
878 }
879 
880 /**
881  * sgx_enclave_modify_types() - Modify type of SGX enclave pages
882  * @encl:	Enclave to which the pages belong.
883  * @modt:	Checked parameters from user about which pages need modifying
884  *              and their new page type.
885  *
886  * Return:
887  * - 0:		Success
888  * - -errno:	Otherwise
889  */
sgx_enclave_modify_types(struct sgx_encl * encl,struct sgx_enclave_modify_types * modt)890 static long sgx_enclave_modify_types(struct sgx_encl *encl,
891 				     struct sgx_enclave_modify_types *modt)
892 {
893 	unsigned long max_prot_restore;
894 	enum sgx_page_type page_type;
895 	struct sgx_encl_page *entry;
896 	struct sgx_secinfo secinfo;
897 	unsigned long prot;
898 	unsigned long addr;
899 	unsigned long c;
900 	void *epc_virt;
901 	int ret;
902 
903 	page_type = modt->page_type & SGX_PAGE_TYPE_MASK;
904 
905 	/*
906 	 * The only new page types allowed by hardware are PT_TCS and PT_TRIM.
907 	 */
908 	if (page_type != SGX_PAGE_TYPE_TCS && page_type != SGX_PAGE_TYPE_TRIM)
909 		return -EINVAL;
910 
911 	memset(&secinfo, 0, sizeof(secinfo));
912 
913 	secinfo.flags = page_type << 8;
914 
915 	for (c = 0 ; c < modt->length; c += PAGE_SIZE) {
916 		addr = encl->base + modt->offset + c;
917 
918 		sgx_reclaim_direct();
919 
920 		mutex_lock(&encl->lock);
921 
922 		entry = sgx_encl_load_page(encl, addr);
923 		if (IS_ERR(entry)) {
924 			ret = PTR_ERR(entry) == -EBUSY ? -EAGAIN : -EFAULT;
925 			goto out_unlock;
926 		}
927 
928 		/*
929 		 * Borrow the logic from the Intel SDM. Regular pages
930 		 * (SGX_PAGE_TYPE_REG) can change type to SGX_PAGE_TYPE_TCS
931 		 * or SGX_PAGE_TYPE_TRIM but TCS pages can only be trimmed.
932 		 * CET pages not supported yet.
933 		 */
934 		if (!(entry->type == SGX_PAGE_TYPE_REG ||
935 		      (entry->type == SGX_PAGE_TYPE_TCS &&
936 		       page_type == SGX_PAGE_TYPE_TRIM))) {
937 			ret = -EINVAL;
938 			goto out_unlock;
939 		}
940 
941 		max_prot_restore = entry->vm_max_prot_bits;
942 
943 		/*
944 		 * Once a regular page becomes a TCS page it cannot be
945 		 * changed back. So the maximum allowed protection reflects
946 		 * the TCS page that is always RW from kernel perspective but
947 		 * will be inaccessible from within enclave. Before doing
948 		 * so, do make sure that the new page type continues to
949 		 * respect the originally vetted page permissions.
950 		 */
951 		if (entry->type == SGX_PAGE_TYPE_REG &&
952 		    page_type == SGX_PAGE_TYPE_TCS) {
953 			if (~entry->vm_max_prot_bits & (VM_READ | VM_WRITE)) {
954 				ret = -EPERM;
955 				goto out_unlock;
956 			}
957 			prot = PROT_READ | PROT_WRITE;
958 			entry->vm_max_prot_bits = calc_vm_prot_bits(prot, 0);
959 
960 			/*
961 			 * Prevent page from being reclaimed while mutex
962 			 * is released.
963 			 */
964 			if (sgx_unmark_page_reclaimable(entry->epc_page)) {
965 				ret = -EAGAIN;
966 				goto out_entry_changed;
967 			}
968 
969 			/*
970 			 * Do not keep encl->lock because of dependency on
971 			 * mmap_lock acquired in sgx_zap_enclave_ptes().
972 			 */
973 			mutex_unlock(&encl->lock);
974 
975 			sgx_zap_enclave_ptes(encl, addr);
976 
977 			mutex_lock(&encl->lock);
978 
979 			sgx_mark_page_reclaimable(entry->epc_page);
980 		}
981 
982 		/* Change EPC type */
983 		epc_virt = sgx_get_epc_virt_addr(entry->epc_page);
984 		ret = __emodt(&secinfo, epc_virt);
985 		if (encls_faulted(ret)) {
986 			/*
987 			 * All possible faults should be avoidable:
988 			 * parameters have been checked, will only change
989 			 * valid page types, and no concurrent
990 			 * SGX1/SGX2 ENCLS instructions since these are
991 			 * protected with mutex.
992 			 */
993 			pr_err_once("EMODT encountered exception %d\n",
994 				    ENCLS_TRAPNR(ret));
995 			ret = -EFAULT;
996 			goto out_entry_changed;
997 		}
998 		if (encls_failed(ret)) {
999 			modt->result = ret;
1000 			ret = -EFAULT;
1001 			goto out_entry_changed;
1002 		}
1003 
1004 		ret = sgx_enclave_etrack(encl);
1005 		if (ret) {
1006 			ret = -EFAULT;
1007 			goto out_unlock;
1008 		}
1009 
1010 		entry->type = page_type;
1011 
1012 		mutex_unlock(&encl->lock);
1013 	}
1014 
1015 	ret = 0;
1016 	goto out;
1017 
1018 out_entry_changed:
1019 	entry->vm_max_prot_bits = max_prot_restore;
1020 out_unlock:
1021 	mutex_unlock(&encl->lock);
1022 out:
1023 	modt->count = c;
1024 
1025 	return ret;
1026 }
1027 
1028 /**
1029  * sgx_ioc_enclave_modify_types() - handler for %SGX_IOC_ENCLAVE_MODIFY_TYPES
1030  * @encl:	an enclave pointer
1031  * @arg:	userspace pointer to a &struct sgx_enclave_modify_types instance
1032  *
1033  * Ability to change the enclave page type supports the following use cases:
1034  *
1035  * * It is possible to add TCS pages to an enclave by changing the type of
1036  *   regular pages (%SGX_PAGE_TYPE_REG) to TCS (%SGX_PAGE_TYPE_TCS) pages.
1037  *   With this support the number of threads supported by an initialized
1038  *   enclave can be increased dynamically.
1039  *
1040  * * Regular or TCS pages can dynamically be removed from an initialized
1041  *   enclave by changing the page type to %SGX_PAGE_TYPE_TRIM. Changing the
1042  *   page type to %SGX_PAGE_TYPE_TRIM marks the page for removal with actual
1043  *   removal done by handler of %SGX_IOC_ENCLAVE_REMOVE_PAGES ioctl() called
1044  *   after ENCLU[EACCEPT] is run on %SGX_PAGE_TYPE_TRIM page from within the
1045  *   enclave.
1046  *
1047  * Return:
1048  * - 0:		Success
1049  * - -errno:	Otherwise
1050  */
sgx_ioc_enclave_modify_types(struct sgx_encl * encl,void __user * arg)1051 static long sgx_ioc_enclave_modify_types(struct sgx_encl *encl,
1052 					 void __user *arg)
1053 {
1054 	struct sgx_enclave_modify_types params;
1055 	long ret;
1056 
1057 	ret = sgx_ioc_sgx2_ready(encl);
1058 	if (ret)
1059 		return ret;
1060 
1061 	if (copy_from_user(&params, arg, sizeof(params)))
1062 		return -EFAULT;
1063 
1064 	if (sgx_validate_offset_length(encl, params.offset, params.length))
1065 		return -EINVAL;
1066 
1067 	if (params.page_type & ~SGX_PAGE_TYPE_MASK)
1068 		return -EINVAL;
1069 
1070 	if (params.result || params.count)
1071 		return -EINVAL;
1072 
1073 	ret = sgx_enclave_modify_types(encl, &params);
1074 
1075 	if (copy_to_user(arg, &params, sizeof(params)))
1076 		return -EFAULT;
1077 
1078 	return ret;
1079 }
1080 
1081 /**
1082  * sgx_encl_remove_pages() - Remove trimmed pages from SGX enclave
1083  * @encl:	Enclave to which the pages belong
1084  * @params:	Checked parameters from user on which pages need to be removed
1085  *
1086  * Return:
1087  * - 0:		Success.
1088  * - -errno:	Otherwise.
1089  */
sgx_encl_remove_pages(struct sgx_encl * encl,struct sgx_enclave_remove_pages * params)1090 static long sgx_encl_remove_pages(struct sgx_encl *encl,
1091 				  struct sgx_enclave_remove_pages *params)
1092 {
1093 	struct sgx_encl_page *entry;
1094 	struct sgx_secinfo secinfo;
1095 	unsigned long addr;
1096 	unsigned long c;
1097 	void *epc_virt;
1098 	int ret;
1099 
1100 	memset(&secinfo, 0, sizeof(secinfo));
1101 	secinfo.flags = SGX_SECINFO_R | SGX_SECINFO_W | SGX_SECINFO_X;
1102 
1103 	for (c = 0 ; c < params->length; c += PAGE_SIZE) {
1104 		addr = encl->base + params->offset + c;
1105 
1106 		sgx_reclaim_direct();
1107 
1108 		mutex_lock(&encl->lock);
1109 
1110 		entry = sgx_encl_load_page(encl, addr);
1111 		if (IS_ERR(entry)) {
1112 			ret = PTR_ERR(entry) == -EBUSY ? -EAGAIN : -EFAULT;
1113 			goto out_unlock;
1114 		}
1115 
1116 		if (entry->type != SGX_PAGE_TYPE_TRIM) {
1117 			ret = -EPERM;
1118 			goto out_unlock;
1119 		}
1120 
1121 		/*
1122 		 * ENCLS[EMODPR] is a no-op instruction used to inform if
1123 		 * ENCLU[EACCEPT] was run from within the enclave. If
1124 		 * ENCLS[EMODPR] is run with RWX on a trimmed page that is
1125 		 * not yet accepted then it will return
1126 		 * %SGX_PAGE_NOT_MODIFIABLE, after the trimmed page is
1127 		 * accepted the instruction will encounter a page fault.
1128 		 */
1129 		epc_virt = sgx_get_epc_virt_addr(entry->epc_page);
1130 		ret = __emodpr(&secinfo, epc_virt);
1131 		if (!encls_faulted(ret) || ENCLS_TRAPNR(ret) != X86_TRAP_PF) {
1132 			ret = -EPERM;
1133 			goto out_unlock;
1134 		}
1135 
1136 		if (sgx_unmark_page_reclaimable(entry->epc_page)) {
1137 			ret = -EBUSY;
1138 			goto out_unlock;
1139 		}
1140 
1141 		/*
1142 		 * Do not keep encl->lock because of dependency on
1143 		 * mmap_lock acquired in sgx_zap_enclave_ptes().
1144 		 */
1145 		mutex_unlock(&encl->lock);
1146 
1147 		sgx_zap_enclave_ptes(encl, addr);
1148 
1149 		mutex_lock(&encl->lock);
1150 
1151 		sgx_encl_free_epc_page(entry->epc_page);
1152 		encl->secs_child_cnt--;
1153 		entry->epc_page = NULL;
1154 		xa_erase(&encl->page_array, PFN_DOWN(entry->desc));
1155 		sgx_encl_shrink(encl, NULL);
1156 		kfree(entry);
1157 
1158 		mutex_unlock(&encl->lock);
1159 	}
1160 
1161 	ret = 0;
1162 	goto out;
1163 
1164 out_unlock:
1165 	mutex_unlock(&encl->lock);
1166 out:
1167 	params->count = c;
1168 
1169 	return ret;
1170 }
1171 
1172 /**
1173  * sgx_ioc_enclave_remove_pages() - handler for %SGX_IOC_ENCLAVE_REMOVE_PAGES
1174  * @encl:	an enclave pointer
1175  * @arg:	userspace pointer to &struct sgx_enclave_remove_pages instance
1176  *
1177  * Final step of the flow removing pages from an initialized enclave. The
1178  * complete flow is:
1179  *
1180  * 1) User changes the type of the pages to be removed to %SGX_PAGE_TYPE_TRIM
1181  *    using the %SGX_IOC_ENCLAVE_MODIFY_TYPES ioctl().
1182  * 2) User approves the page removal by running ENCLU[EACCEPT] from within
1183  *    the enclave.
1184  * 3) User initiates actual page removal using the
1185  *    %SGX_IOC_ENCLAVE_REMOVE_PAGES ioctl() that is handled here.
1186  *
1187  * First remove any page table entries pointing to the page and then proceed
1188  * with the actual removal of the enclave page and data in support of it.
1189  *
1190  * VA pages are not affected by this removal. It is thus possible that the
1191  * enclave may end up with more VA pages than needed to support all its
1192  * pages.
1193  *
1194  * Return:
1195  * - 0:		Success
1196  * - -errno:	Otherwise
1197  */
sgx_ioc_enclave_remove_pages(struct sgx_encl * encl,void __user * arg)1198 static long sgx_ioc_enclave_remove_pages(struct sgx_encl *encl,
1199 					 void __user *arg)
1200 {
1201 	struct sgx_enclave_remove_pages params;
1202 	long ret;
1203 
1204 	ret = sgx_ioc_sgx2_ready(encl);
1205 	if (ret)
1206 		return ret;
1207 
1208 	if (copy_from_user(&params, arg, sizeof(params)))
1209 		return -EFAULT;
1210 
1211 	if (sgx_validate_offset_length(encl, params.offset, params.length))
1212 		return -EINVAL;
1213 
1214 	if (params.count)
1215 		return -EINVAL;
1216 
1217 	ret = sgx_encl_remove_pages(encl, &params);
1218 
1219 	if (copy_to_user(arg, &params, sizeof(params)))
1220 		return -EFAULT;
1221 
1222 	return ret;
1223 }
1224 
sgx_ioctl(struct file * filep,unsigned int cmd,unsigned long arg)1225 long sgx_ioctl(struct file *filep, unsigned int cmd, unsigned long arg)
1226 {
1227 	struct sgx_encl *encl = filep->private_data;
1228 	int ret;
1229 
1230 	if (test_and_set_bit(SGX_ENCL_IOCTL, &encl->flags))
1231 		return -EBUSY;
1232 
1233 	switch (cmd) {
1234 	case SGX_IOC_ENCLAVE_CREATE:
1235 		ret = sgx_ioc_enclave_create(encl, (void __user *)arg);
1236 		break;
1237 	case SGX_IOC_ENCLAVE_ADD_PAGES:
1238 		ret = sgx_ioc_enclave_add_pages(encl, (void __user *)arg);
1239 		break;
1240 	case SGX_IOC_ENCLAVE_INIT:
1241 		ret = sgx_ioc_enclave_init(encl, (void __user *)arg);
1242 		break;
1243 	case SGX_IOC_ENCLAVE_PROVISION:
1244 		ret = sgx_ioc_enclave_provision(encl, (void __user *)arg);
1245 		break;
1246 	case SGX_IOC_ENCLAVE_RESTRICT_PERMISSIONS:
1247 		ret = sgx_ioc_enclave_restrict_permissions(encl,
1248 							   (void __user *)arg);
1249 		break;
1250 	case SGX_IOC_ENCLAVE_MODIFY_TYPES:
1251 		ret = sgx_ioc_enclave_modify_types(encl, (void __user *)arg);
1252 		break;
1253 	case SGX_IOC_ENCLAVE_REMOVE_PAGES:
1254 		ret = sgx_ioc_enclave_remove_pages(encl, (void __user *)arg);
1255 		break;
1256 	default:
1257 		ret = -ENOIOCTLCMD;
1258 		break;
1259 	}
1260 
1261 	clear_bit(SGX_ENCL_IOCTL, &encl->flags);
1262 	return ret;
1263 }
1264