xref: /openbmc/linux/drivers/xen/privcmd.c (revision f9a82c48)
1 /******************************************************************************
2  * privcmd.c
3  *
4  * Interface to privileged domain-0 commands.
5  *
6  * Copyright (c) 2002-2004, K A Fraser, B Dragovic
7  */
8 
9 #define pr_fmt(fmt) "xen:" KBUILD_MODNAME ": " fmt
10 
11 #include <linux/kernel.h>
12 #include <linux/module.h>
13 #include <linux/sched.h>
14 #include <linux/slab.h>
15 #include <linux/string.h>
16 #include <linux/errno.h>
17 #include <linux/mm.h>
18 #include <linux/mman.h>
19 #include <linux/uaccess.h>
20 #include <linux/swap.h>
21 #include <linux/highmem.h>
22 #include <linux/pagemap.h>
23 #include <linux/seq_file.h>
24 #include <linux/miscdevice.h>
25 #include <linux/moduleparam.h>
26 
27 #include <asm/pgalloc.h>
28 #include <asm/pgtable.h>
29 #include <asm/tlb.h>
30 #include <asm/xen/hypervisor.h>
31 #include <asm/xen/hypercall.h>
32 
33 #include <xen/xen.h>
34 #include <xen/privcmd.h>
35 #include <xen/interface/xen.h>
36 #include <xen/interface/memory.h>
37 #include <xen/interface/hvm/dm_op.h>
38 #include <xen/features.h>
39 #include <xen/page.h>
40 #include <xen/xen-ops.h>
41 #include <xen/balloon.h>
42 
43 #include "privcmd.h"
44 
45 MODULE_LICENSE("GPL");
46 
47 #define PRIV_VMA_LOCKED ((void *)1)
48 
49 static unsigned int privcmd_dm_op_max_num = 16;
50 module_param_named(dm_op_max_nr_bufs, privcmd_dm_op_max_num, uint, 0644);
51 MODULE_PARM_DESC(dm_op_max_nr_bufs,
52 		 "Maximum number of buffers per dm_op hypercall");
53 
54 static unsigned int privcmd_dm_op_buf_max_size = 4096;
55 module_param_named(dm_op_buf_max_size, privcmd_dm_op_buf_max_size, uint,
56 		   0644);
57 MODULE_PARM_DESC(dm_op_buf_max_size,
58 		 "Maximum size of a dm_op hypercall buffer");
59 
60 struct privcmd_data {
61 	domid_t domid;
62 };
63 
64 static int privcmd_vma_range_is_mapped(
65                struct vm_area_struct *vma,
66                unsigned long addr,
67                unsigned long nr_pages);
68 
69 static long privcmd_ioctl_hypercall(struct file *file, void __user *udata)
70 {
71 	struct privcmd_data *data = file->private_data;
72 	struct privcmd_hypercall hypercall;
73 	long ret;
74 
75 	/* Disallow arbitrary hypercalls if restricted */
76 	if (data->domid != DOMID_INVALID)
77 		return -EPERM;
78 
79 	if (copy_from_user(&hypercall, udata, sizeof(hypercall)))
80 		return -EFAULT;
81 
82 	xen_preemptible_hcall_begin();
83 	ret = privcmd_call(hypercall.op,
84 			   hypercall.arg[0], hypercall.arg[1],
85 			   hypercall.arg[2], hypercall.arg[3],
86 			   hypercall.arg[4]);
87 	xen_preemptible_hcall_end();
88 
89 	return ret;
90 }
91 
92 static void free_page_list(struct list_head *pages)
93 {
94 	struct page *p, *n;
95 
96 	list_for_each_entry_safe(p, n, pages, lru)
97 		__free_page(p);
98 
99 	INIT_LIST_HEAD(pages);
100 }
101 
102 /*
103  * Given an array of items in userspace, return a list of pages
104  * containing the data.  If copying fails, either because of memory
105  * allocation failure or a problem reading user memory, return an
106  * error code; its up to the caller to dispose of any partial list.
107  */
108 static int gather_array(struct list_head *pagelist,
109 			unsigned nelem, size_t size,
110 			const void __user *data)
111 {
112 	unsigned pageidx;
113 	void *pagedata;
114 	int ret;
115 
116 	if (size > PAGE_SIZE)
117 		return 0;
118 
119 	pageidx = PAGE_SIZE;
120 	pagedata = NULL;	/* quiet, gcc */
121 	while (nelem--) {
122 		if (pageidx > PAGE_SIZE-size) {
123 			struct page *page = alloc_page(GFP_KERNEL);
124 
125 			ret = -ENOMEM;
126 			if (page == NULL)
127 				goto fail;
128 
129 			pagedata = page_address(page);
130 
131 			list_add_tail(&page->lru, pagelist);
132 			pageidx = 0;
133 		}
134 
135 		ret = -EFAULT;
136 		if (copy_from_user(pagedata + pageidx, data, size))
137 			goto fail;
138 
139 		data += size;
140 		pageidx += size;
141 	}
142 
143 	ret = 0;
144 
145 fail:
146 	return ret;
147 }
148 
149 /*
150  * Call function "fn" on each element of the array fragmented
151  * over a list of pages.
152  */
153 static int traverse_pages(unsigned nelem, size_t size,
154 			  struct list_head *pos,
155 			  int (*fn)(void *data, void *state),
156 			  void *state)
157 {
158 	void *pagedata;
159 	unsigned pageidx;
160 	int ret = 0;
161 
162 	BUG_ON(size > PAGE_SIZE);
163 
164 	pageidx = PAGE_SIZE;
165 	pagedata = NULL;	/* hush, gcc */
166 
167 	while (nelem--) {
168 		if (pageidx > PAGE_SIZE-size) {
169 			struct page *page;
170 			pos = pos->next;
171 			page = list_entry(pos, struct page, lru);
172 			pagedata = page_address(page);
173 			pageidx = 0;
174 		}
175 
176 		ret = (*fn)(pagedata + pageidx, state);
177 		if (ret)
178 			break;
179 		pageidx += size;
180 	}
181 
182 	return ret;
183 }
184 
185 /*
186  * Similar to traverse_pages, but use each page as a "block" of
187  * data to be processed as one unit.
188  */
189 static int traverse_pages_block(unsigned nelem, size_t size,
190 				struct list_head *pos,
191 				int (*fn)(void *data, int nr, void *state),
192 				void *state)
193 {
194 	void *pagedata;
195 	int ret = 0;
196 
197 	BUG_ON(size > PAGE_SIZE);
198 
199 	while (nelem) {
200 		int nr = (PAGE_SIZE/size);
201 		struct page *page;
202 		if (nr > nelem)
203 			nr = nelem;
204 		pos = pos->next;
205 		page = list_entry(pos, struct page, lru);
206 		pagedata = page_address(page);
207 		ret = (*fn)(pagedata, nr, state);
208 		if (ret)
209 			break;
210 		nelem -= nr;
211 	}
212 
213 	return ret;
214 }
215 
216 struct mmap_gfn_state {
217 	unsigned long va;
218 	struct vm_area_struct *vma;
219 	domid_t domain;
220 };
221 
222 static int mmap_gfn_range(void *data, void *state)
223 {
224 	struct privcmd_mmap_entry *msg = data;
225 	struct mmap_gfn_state *st = state;
226 	struct vm_area_struct *vma = st->vma;
227 	int rc;
228 
229 	/* Do not allow range to wrap the address space. */
230 	if ((msg->npages > (LONG_MAX >> PAGE_SHIFT)) ||
231 	    ((unsigned long)(msg->npages << PAGE_SHIFT) >= -st->va))
232 		return -EINVAL;
233 
234 	/* Range chunks must be contiguous in va space. */
235 	if ((msg->va != st->va) ||
236 	    ((msg->va+(msg->npages<<PAGE_SHIFT)) > vma->vm_end))
237 		return -EINVAL;
238 
239 	rc = xen_remap_domain_gfn_range(vma,
240 					msg->va & PAGE_MASK,
241 					msg->mfn, msg->npages,
242 					vma->vm_page_prot,
243 					st->domain, NULL);
244 	if (rc < 0)
245 		return rc;
246 
247 	st->va += msg->npages << PAGE_SHIFT;
248 
249 	return 0;
250 }
251 
252 static long privcmd_ioctl_mmap(struct file *file, void __user *udata)
253 {
254 	struct privcmd_data *data = file->private_data;
255 	struct privcmd_mmap mmapcmd;
256 	struct mm_struct *mm = current->mm;
257 	struct vm_area_struct *vma;
258 	int rc;
259 	LIST_HEAD(pagelist);
260 	struct mmap_gfn_state state;
261 
262 	/* We only support privcmd_ioctl_mmap_batch for auto translated. */
263 	if (xen_feature(XENFEAT_auto_translated_physmap))
264 		return -ENOSYS;
265 
266 	if (copy_from_user(&mmapcmd, udata, sizeof(mmapcmd)))
267 		return -EFAULT;
268 
269 	/* If restriction is in place, check the domid matches */
270 	if (data->domid != DOMID_INVALID && data->domid != mmapcmd.dom)
271 		return -EPERM;
272 
273 	rc = gather_array(&pagelist,
274 			  mmapcmd.num, sizeof(struct privcmd_mmap_entry),
275 			  mmapcmd.entry);
276 
277 	if (rc || list_empty(&pagelist))
278 		goto out;
279 
280 	down_write(&mm->mmap_sem);
281 
282 	{
283 		struct page *page = list_first_entry(&pagelist,
284 						     struct page, lru);
285 		struct privcmd_mmap_entry *msg = page_address(page);
286 
287 		vma = find_vma(mm, msg->va);
288 		rc = -EINVAL;
289 
290 		if (!vma || (msg->va != vma->vm_start) || vma->vm_private_data)
291 			goto out_up;
292 		vma->vm_private_data = PRIV_VMA_LOCKED;
293 	}
294 
295 	state.va = vma->vm_start;
296 	state.vma = vma;
297 	state.domain = mmapcmd.dom;
298 
299 	rc = traverse_pages(mmapcmd.num, sizeof(struct privcmd_mmap_entry),
300 			    &pagelist,
301 			    mmap_gfn_range, &state);
302 
303 
304 out_up:
305 	up_write(&mm->mmap_sem);
306 
307 out:
308 	free_page_list(&pagelist);
309 
310 	return rc;
311 }
312 
313 struct mmap_batch_state {
314 	domid_t domain;
315 	unsigned long va;
316 	struct vm_area_struct *vma;
317 	int index;
318 	/* A tristate:
319 	 *      0 for no errors
320 	 *      1 if at least one error has happened (and no
321 	 *          -ENOENT errors have happened)
322 	 *      -ENOENT if at least 1 -ENOENT has happened.
323 	 */
324 	int global_error;
325 	int version;
326 
327 	/* User-space gfn array to store errors in the second pass for V1. */
328 	xen_pfn_t __user *user_gfn;
329 	/* User-space int array to store errors in the second pass for V2. */
330 	int __user *user_err;
331 };
332 
333 /* auto translated dom0 note: if domU being created is PV, then gfn is
334  * mfn(addr on bus). If it's auto xlated, then gfn is pfn (input to HAP).
335  */
336 static int mmap_batch_fn(void *data, int nr, void *state)
337 {
338 	xen_pfn_t *gfnp = data;
339 	struct mmap_batch_state *st = state;
340 	struct vm_area_struct *vma = st->vma;
341 	struct page **pages = vma->vm_private_data;
342 	struct page **cur_pages = NULL;
343 	int ret;
344 
345 	if (xen_feature(XENFEAT_auto_translated_physmap))
346 		cur_pages = &pages[st->index];
347 
348 	BUG_ON(nr < 0);
349 	ret = xen_remap_domain_gfn_array(st->vma, st->va & PAGE_MASK, gfnp, nr,
350 					 (int *)gfnp, st->vma->vm_page_prot,
351 					 st->domain, cur_pages);
352 
353 	/* Adjust the global_error? */
354 	if (ret != nr) {
355 		if (ret == -ENOENT)
356 			st->global_error = -ENOENT;
357 		else {
358 			/* Record that at least one error has happened. */
359 			if (st->global_error == 0)
360 				st->global_error = 1;
361 		}
362 	}
363 	st->va += XEN_PAGE_SIZE * nr;
364 	st->index += nr / XEN_PFN_PER_PAGE;
365 
366 	return 0;
367 }
368 
369 static int mmap_return_error(int err, struct mmap_batch_state *st)
370 {
371 	int ret;
372 
373 	if (st->version == 1) {
374 		if (err) {
375 			xen_pfn_t gfn;
376 
377 			ret = get_user(gfn, st->user_gfn);
378 			if (ret < 0)
379 				return ret;
380 			/*
381 			 * V1 encodes the error codes in the 32bit top
382 			 * nibble of the gfn (with its known
383 			 * limitations vis-a-vis 64 bit callers).
384 			 */
385 			gfn |= (err == -ENOENT) ?
386 				PRIVCMD_MMAPBATCH_PAGED_ERROR :
387 				PRIVCMD_MMAPBATCH_MFN_ERROR;
388 			return __put_user(gfn, st->user_gfn++);
389 		} else
390 			st->user_gfn++;
391 	} else { /* st->version == 2 */
392 		if (err)
393 			return __put_user(err, st->user_err++);
394 		else
395 			st->user_err++;
396 	}
397 
398 	return 0;
399 }
400 
401 static int mmap_return_errors(void *data, int nr, void *state)
402 {
403 	struct mmap_batch_state *st = state;
404 	int *errs = data;
405 	int i;
406 	int ret;
407 
408 	for (i = 0; i < nr; i++) {
409 		ret = mmap_return_error(errs[i], st);
410 		if (ret < 0)
411 			return ret;
412 	}
413 	return 0;
414 }
415 
416 /* Allocate pfns that are then mapped with gfns from foreign domid. Update
417  * the vma with the page info to use later.
418  * Returns: 0 if success, otherwise -errno
419  */
420 static int alloc_empty_pages(struct vm_area_struct *vma, int numpgs)
421 {
422 	int rc;
423 	struct page **pages;
424 
425 	pages = kcalloc(numpgs, sizeof(pages[0]), GFP_KERNEL);
426 	if (pages == NULL)
427 		return -ENOMEM;
428 
429 	rc = alloc_xenballooned_pages(numpgs, pages);
430 	if (rc != 0) {
431 		pr_warn("%s Could not alloc %d pfns rc:%d\n", __func__,
432 			numpgs, rc);
433 		kfree(pages);
434 		return -ENOMEM;
435 	}
436 	BUG_ON(vma->vm_private_data != NULL);
437 	vma->vm_private_data = pages;
438 
439 	return 0;
440 }
441 
442 static const struct vm_operations_struct privcmd_vm_ops;
443 
444 static long privcmd_ioctl_mmap_batch(
445 	struct file *file, void __user *udata, int version)
446 {
447 	struct privcmd_data *data = file->private_data;
448 	int ret;
449 	struct privcmd_mmapbatch_v2 m;
450 	struct mm_struct *mm = current->mm;
451 	struct vm_area_struct *vma;
452 	unsigned long nr_pages;
453 	LIST_HEAD(pagelist);
454 	struct mmap_batch_state state;
455 
456 	switch (version) {
457 	case 1:
458 		if (copy_from_user(&m, udata, sizeof(struct privcmd_mmapbatch)))
459 			return -EFAULT;
460 		/* Returns per-frame error in m.arr. */
461 		m.err = NULL;
462 		if (!access_ok(m.arr, m.num * sizeof(*m.arr)))
463 			return -EFAULT;
464 		break;
465 	case 2:
466 		if (copy_from_user(&m, udata, sizeof(struct privcmd_mmapbatch_v2)))
467 			return -EFAULT;
468 		/* Returns per-frame error code in m.err. */
469 		if (!access_ok(m.err, m.num * (sizeof(*m.err))))
470 			return -EFAULT;
471 		break;
472 	default:
473 		return -EINVAL;
474 	}
475 
476 	/* If restriction is in place, check the domid matches */
477 	if (data->domid != DOMID_INVALID && data->domid != m.dom)
478 		return -EPERM;
479 
480 	nr_pages = DIV_ROUND_UP(m.num, XEN_PFN_PER_PAGE);
481 	if ((m.num <= 0) || (nr_pages > (LONG_MAX >> PAGE_SHIFT)))
482 		return -EINVAL;
483 
484 	ret = gather_array(&pagelist, m.num, sizeof(xen_pfn_t), m.arr);
485 
486 	if (ret)
487 		goto out;
488 	if (list_empty(&pagelist)) {
489 		ret = -EINVAL;
490 		goto out;
491 	}
492 
493 	if (version == 2) {
494 		/* Zero error array now to only copy back actual errors. */
495 		if (clear_user(m.err, sizeof(int) * m.num)) {
496 			ret = -EFAULT;
497 			goto out;
498 		}
499 	}
500 
501 	down_write(&mm->mmap_sem);
502 
503 	vma = find_vma(mm, m.addr);
504 	if (!vma ||
505 	    vma->vm_ops != &privcmd_vm_ops) {
506 		ret = -EINVAL;
507 		goto out_unlock;
508 	}
509 
510 	/*
511 	 * Caller must either:
512 	 *
513 	 * Map the whole VMA range, which will also allocate all the
514 	 * pages required for the auto_translated_physmap case.
515 	 *
516 	 * Or
517 	 *
518 	 * Map unmapped holes left from a previous map attempt (e.g.,
519 	 * because those foreign frames were previously paged out).
520 	 */
521 	if (vma->vm_private_data == NULL) {
522 		if (m.addr != vma->vm_start ||
523 		    m.addr + (nr_pages << PAGE_SHIFT) != vma->vm_end) {
524 			ret = -EINVAL;
525 			goto out_unlock;
526 		}
527 		if (xen_feature(XENFEAT_auto_translated_physmap)) {
528 			ret = alloc_empty_pages(vma, nr_pages);
529 			if (ret < 0)
530 				goto out_unlock;
531 		} else
532 			vma->vm_private_data = PRIV_VMA_LOCKED;
533 	} else {
534 		if (m.addr < vma->vm_start ||
535 		    m.addr + (nr_pages << PAGE_SHIFT) > vma->vm_end) {
536 			ret = -EINVAL;
537 			goto out_unlock;
538 		}
539 		if (privcmd_vma_range_is_mapped(vma, m.addr, nr_pages)) {
540 			ret = -EINVAL;
541 			goto out_unlock;
542 		}
543 	}
544 
545 	state.domain        = m.dom;
546 	state.vma           = vma;
547 	state.va            = m.addr;
548 	state.index         = 0;
549 	state.global_error  = 0;
550 	state.version       = version;
551 
552 	BUILD_BUG_ON(((PAGE_SIZE / sizeof(xen_pfn_t)) % XEN_PFN_PER_PAGE) != 0);
553 	/* mmap_batch_fn guarantees ret == 0 */
554 	BUG_ON(traverse_pages_block(m.num, sizeof(xen_pfn_t),
555 				    &pagelist, mmap_batch_fn, &state));
556 
557 	up_write(&mm->mmap_sem);
558 
559 	if (state.global_error) {
560 		/* Write back errors in second pass. */
561 		state.user_gfn = (xen_pfn_t *)m.arr;
562 		state.user_err = m.err;
563 		ret = traverse_pages_block(m.num, sizeof(xen_pfn_t),
564 					   &pagelist, mmap_return_errors, &state);
565 	} else
566 		ret = 0;
567 
568 	/* If we have not had any EFAULT-like global errors then set the global
569 	 * error to -ENOENT if necessary. */
570 	if ((ret == 0) && (state.global_error == -ENOENT))
571 		ret = -ENOENT;
572 
573 out:
574 	free_page_list(&pagelist);
575 	return ret;
576 
577 out_unlock:
578 	up_write(&mm->mmap_sem);
579 	goto out;
580 }
581 
582 static int lock_pages(
583 	struct privcmd_dm_op_buf kbufs[], unsigned int num,
584 	struct page *pages[], unsigned int nr_pages)
585 {
586 	unsigned int i;
587 
588 	for (i = 0; i < num; i++) {
589 		unsigned int requested;
590 		int pinned;
591 
592 		requested = DIV_ROUND_UP(
593 			offset_in_page(kbufs[i].uptr) + kbufs[i].size,
594 			PAGE_SIZE);
595 		if (requested > nr_pages)
596 			return -ENOSPC;
597 
598 		pinned = get_user_pages_fast(
599 			(unsigned long) kbufs[i].uptr,
600 			requested, FOLL_WRITE, pages);
601 		if (pinned < 0)
602 			return pinned;
603 
604 		nr_pages -= pinned;
605 		pages += pinned;
606 	}
607 
608 	return 0;
609 }
610 
611 static void unlock_pages(struct page *pages[], unsigned int nr_pages)
612 {
613 	unsigned int i;
614 
615 	if (!pages)
616 		return;
617 
618 	for (i = 0; i < nr_pages; i++) {
619 		if (pages[i])
620 			put_page(pages[i]);
621 	}
622 }
623 
624 static long privcmd_ioctl_dm_op(struct file *file, void __user *udata)
625 {
626 	struct privcmd_data *data = file->private_data;
627 	struct privcmd_dm_op kdata;
628 	struct privcmd_dm_op_buf *kbufs;
629 	unsigned int nr_pages = 0;
630 	struct page **pages = NULL;
631 	struct xen_dm_op_buf *xbufs = NULL;
632 	unsigned int i;
633 	long rc;
634 
635 	if (copy_from_user(&kdata, udata, sizeof(kdata)))
636 		return -EFAULT;
637 
638 	/* If restriction is in place, check the domid matches */
639 	if (data->domid != DOMID_INVALID && data->domid != kdata.dom)
640 		return -EPERM;
641 
642 	if (kdata.num == 0)
643 		return 0;
644 
645 	if (kdata.num > privcmd_dm_op_max_num)
646 		return -E2BIG;
647 
648 	kbufs = kcalloc(kdata.num, sizeof(*kbufs), GFP_KERNEL);
649 	if (!kbufs)
650 		return -ENOMEM;
651 
652 	if (copy_from_user(kbufs, kdata.ubufs,
653 			   sizeof(*kbufs) * kdata.num)) {
654 		rc = -EFAULT;
655 		goto out;
656 	}
657 
658 	for (i = 0; i < kdata.num; i++) {
659 		if (kbufs[i].size > privcmd_dm_op_buf_max_size) {
660 			rc = -E2BIG;
661 			goto out;
662 		}
663 
664 		if (!access_ok(kbufs[i].uptr,
665 			       kbufs[i].size)) {
666 			rc = -EFAULT;
667 			goto out;
668 		}
669 
670 		nr_pages += DIV_ROUND_UP(
671 			offset_in_page(kbufs[i].uptr) + kbufs[i].size,
672 			PAGE_SIZE);
673 	}
674 
675 	pages = kcalloc(nr_pages, sizeof(*pages), GFP_KERNEL);
676 	if (!pages) {
677 		rc = -ENOMEM;
678 		goto out;
679 	}
680 
681 	xbufs = kcalloc(kdata.num, sizeof(*xbufs), GFP_KERNEL);
682 	if (!xbufs) {
683 		rc = -ENOMEM;
684 		goto out;
685 	}
686 
687 	rc = lock_pages(kbufs, kdata.num, pages, nr_pages);
688 	if (rc)
689 		goto out;
690 
691 	for (i = 0; i < kdata.num; i++) {
692 		set_xen_guest_handle(xbufs[i].h, kbufs[i].uptr);
693 		xbufs[i].size = kbufs[i].size;
694 	}
695 
696 	xen_preemptible_hcall_begin();
697 	rc = HYPERVISOR_dm_op(kdata.dom, kdata.num, xbufs);
698 	xen_preemptible_hcall_end();
699 
700 out:
701 	unlock_pages(pages, nr_pages);
702 	kfree(xbufs);
703 	kfree(pages);
704 	kfree(kbufs);
705 
706 	return rc;
707 }
708 
709 static long privcmd_ioctl_restrict(struct file *file, void __user *udata)
710 {
711 	struct privcmd_data *data = file->private_data;
712 	domid_t dom;
713 
714 	if (copy_from_user(&dom, udata, sizeof(dom)))
715 		return -EFAULT;
716 
717 	/* Set restriction to the specified domain, or check it matches */
718 	if (data->domid == DOMID_INVALID)
719 		data->domid = dom;
720 	else if (data->domid != dom)
721 		return -EINVAL;
722 
723 	return 0;
724 }
725 
726 struct remap_pfn {
727 	struct mm_struct *mm;
728 	struct page **pages;
729 	pgprot_t prot;
730 	unsigned long i;
731 };
732 
733 static int remap_pfn_fn(pte_t *ptep, pgtable_t token, unsigned long addr,
734 			void *data)
735 {
736 	struct remap_pfn *r = data;
737 	struct page *page = r->pages[r->i];
738 	pte_t pte = pte_mkspecial(pfn_pte(page_to_pfn(page), r->prot));
739 
740 	set_pte_at(r->mm, addr, ptep, pte);
741 	r->i++;
742 
743 	return 0;
744 }
745 
746 static long privcmd_ioctl_mmap_resource(struct file *file, void __user *udata)
747 {
748 	struct privcmd_data *data = file->private_data;
749 	struct mm_struct *mm = current->mm;
750 	struct vm_area_struct *vma;
751 	struct privcmd_mmap_resource kdata;
752 	xen_pfn_t *pfns = NULL;
753 	struct xen_mem_acquire_resource xdata;
754 	int rc;
755 
756 	if (copy_from_user(&kdata, udata, sizeof(kdata)))
757 		return -EFAULT;
758 
759 	/* If restriction is in place, check the domid matches */
760 	if (data->domid != DOMID_INVALID && data->domid != kdata.dom)
761 		return -EPERM;
762 
763 	down_write(&mm->mmap_sem);
764 
765 	vma = find_vma(mm, kdata.addr);
766 	if (!vma || vma->vm_ops != &privcmd_vm_ops) {
767 		rc = -EINVAL;
768 		goto out;
769 	}
770 
771 	pfns = kcalloc(kdata.num, sizeof(*pfns), GFP_KERNEL);
772 	if (!pfns) {
773 		rc = -ENOMEM;
774 		goto out;
775 	}
776 
777 	if (xen_feature(XENFEAT_auto_translated_physmap)) {
778 		unsigned int nr = DIV_ROUND_UP(kdata.num, XEN_PFN_PER_PAGE);
779 		struct page **pages;
780 		unsigned int i;
781 
782 		rc = alloc_empty_pages(vma, nr);
783 		if (rc < 0)
784 			goto out;
785 
786 		pages = vma->vm_private_data;
787 		for (i = 0; i < kdata.num; i++) {
788 			xen_pfn_t pfn =
789 				page_to_xen_pfn(pages[i / XEN_PFN_PER_PAGE]);
790 
791 			pfns[i] = pfn + (i % XEN_PFN_PER_PAGE);
792 		}
793 	} else
794 		vma->vm_private_data = PRIV_VMA_LOCKED;
795 
796 	memset(&xdata, 0, sizeof(xdata));
797 	xdata.domid = kdata.dom;
798 	xdata.type = kdata.type;
799 	xdata.id = kdata.id;
800 	xdata.frame = kdata.idx;
801 	xdata.nr_frames = kdata.num;
802 	set_xen_guest_handle(xdata.frame_list, pfns);
803 
804 	xen_preemptible_hcall_begin();
805 	rc = HYPERVISOR_memory_op(XENMEM_acquire_resource, &xdata);
806 	xen_preemptible_hcall_end();
807 
808 	if (rc)
809 		goto out;
810 
811 	if (xen_feature(XENFEAT_auto_translated_physmap)) {
812 		struct remap_pfn r = {
813 			.mm = vma->vm_mm,
814 			.pages = vma->vm_private_data,
815 			.prot = vma->vm_page_prot,
816 		};
817 
818 		rc = apply_to_page_range(r.mm, kdata.addr,
819 					 kdata.num << PAGE_SHIFT,
820 					 remap_pfn_fn, &r);
821 	} else {
822 		unsigned int domid =
823 			(xdata.flags & XENMEM_rsrc_acq_caller_owned) ?
824 			DOMID_SELF : kdata.dom;
825 		int num;
826 
827 		num = xen_remap_domain_mfn_array(vma,
828 						 kdata.addr & PAGE_MASK,
829 						 pfns, kdata.num, (int *)pfns,
830 						 vma->vm_page_prot,
831 						 domid,
832 						 vma->vm_private_data);
833 		if (num < 0)
834 			rc = num;
835 		else if (num != kdata.num) {
836 			unsigned int i;
837 
838 			for (i = 0; i < num; i++) {
839 				rc = pfns[i];
840 				if (rc < 0)
841 					break;
842 			}
843 		} else
844 			rc = 0;
845 	}
846 
847 out:
848 	up_write(&mm->mmap_sem);
849 	kfree(pfns);
850 
851 	return rc;
852 }
853 
854 static long privcmd_ioctl(struct file *file,
855 			  unsigned int cmd, unsigned long data)
856 {
857 	int ret = -ENOTTY;
858 	void __user *udata = (void __user *) data;
859 
860 	switch (cmd) {
861 	case IOCTL_PRIVCMD_HYPERCALL:
862 		ret = privcmd_ioctl_hypercall(file, udata);
863 		break;
864 
865 	case IOCTL_PRIVCMD_MMAP:
866 		ret = privcmd_ioctl_mmap(file, udata);
867 		break;
868 
869 	case IOCTL_PRIVCMD_MMAPBATCH:
870 		ret = privcmd_ioctl_mmap_batch(file, udata, 1);
871 		break;
872 
873 	case IOCTL_PRIVCMD_MMAPBATCH_V2:
874 		ret = privcmd_ioctl_mmap_batch(file, udata, 2);
875 		break;
876 
877 	case IOCTL_PRIVCMD_DM_OP:
878 		ret = privcmd_ioctl_dm_op(file, udata);
879 		break;
880 
881 	case IOCTL_PRIVCMD_RESTRICT:
882 		ret = privcmd_ioctl_restrict(file, udata);
883 		break;
884 
885 	case IOCTL_PRIVCMD_MMAP_RESOURCE:
886 		ret = privcmd_ioctl_mmap_resource(file, udata);
887 		break;
888 
889 	default:
890 		break;
891 	}
892 
893 	return ret;
894 }
895 
896 static int privcmd_open(struct inode *ino, struct file *file)
897 {
898 	struct privcmd_data *data = kzalloc(sizeof(*data), GFP_KERNEL);
899 
900 	if (!data)
901 		return -ENOMEM;
902 
903 	/* DOMID_INVALID implies no restriction */
904 	data->domid = DOMID_INVALID;
905 
906 	file->private_data = data;
907 	return 0;
908 }
909 
910 static int privcmd_release(struct inode *ino, struct file *file)
911 {
912 	struct privcmd_data *data = file->private_data;
913 
914 	kfree(data);
915 	return 0;
916 }
917 
918 static void privcmd_close(struct vm_area_struct *vma)
919 {
920 	struct page **pages = vma->vm_private_data;
921 	int numpgs = vma_pages(vma);
922 	int numgfns = (vma->vm_end - vma->vm_start) >> XEN_PAGE_SHIFT;
923 	int rc;
924 
925 	if (!xen_feature(XENFEAT_auto_translated_physmap) || !numpgs || !pages)
926 		return;
927 
928 	rc = xen_unmap_domain_gfn_range(vma, numgfns, pages);
929 	if (rc == 0)
930 		free_xenballooned_pages(numpgs, pages);
931 	else
932 		pr_crit("unable to unmap MFN range: leaking %d pages. rc=%d\n",
933 			numpgs, rc);
934 	kfree(pages);
935 }
936 
937 static vm_fault_t privcmd_fault(struct vm_fault *vmf)
938 {
939 	printk(KERN_DEBUG "privcmd_fault: vma=%p %lx-%lx, pgoff=%lx, uv=%p\n",
940 	       vmf->vma, vmf->vma->vm_start, vmf->vma->vm_end,
941 	       vmf->pgoff, (void *)vmf->address);
942 
943 	return VM_FAULT_SIGBUS;
944 }
945 
946 static const struct vm_operations_struct privcmd_vm_ops = {
947 	.close = privcmd_close,
948 	.fault = privcmd_fault
949 };
950 
951 static int privcmd_mmap(struct file *file, struct vm_area_struct *vma)
952 {
953 	/* DONTCOPY is essential for Xen because copy_page_range doesn't know
954 	 * how to recreate these mappings */
955 	vma->vm_flags |= VM_IO | VM_PFNMAP | VM_DONTCOPY |
956 			 VM_DONTEXPAND | VM_DONTDUMP;
957 	vma->vm_ops = &privcmd_vm_ops;
958 	vma->vm_private_data = NULL;
959 
960 	return 0;
961 }
962 
963 /*
964  * For MMAPBATCH*. This allows asserting the singleshot mapping
965  * on a per pfn/pte basis. Mapping calls that fail with ENOENT
966  * can be then retried until success.
967  */
968 static int is_mapped_fn(pte_t *pte, struct page *pmd_page,
969 	                unsigned long addr, void *data)
970 {
971 	return pte_none(*pte) ? 0 : -EBUSY;
972 }
973 
974 static int privcmd_vma_range_is_mapped(
975 	           struct vm_area_struct *vma,
976 	           unsigned long addr,
977 	           unsigned long nr_pages)
978 {
979 	return apply_to_page_range(vma->vm_mm, addr, nr_pages << PAGE_SHIFT,
980 				   is_mapped_fn, NULL) != 0;
981 }
982 
983 const struct file_operations xen_privcmd_fops = {
984 	.owner = THIS_MODULE,
985 	.unlocked_ioctl = privcmd_ioctl,
986 	.open = privcmd_open,
987 	.release = privcmd_release,
988 	.mmap = privcmd_mmap,
989 };
990 EXPORT_SYMBOL_GPL(xen_privcmd_fops);
991 
992 static struct miscdevice privcmd_dev = {
993 	.minor = MISC_DYNAMIC_MINOR,
994 	.name = "xen/privcmd",
995 	.fops = &xen_privcmd_fops,
996 };
997 
998 static int __init privcmd_init(void)
999 {
1000 	int err;
1001 
1002 	if (!xen_domain())
1003 		return -ENODEV;
1004 
1005 	err = misc_register(&privcmd_dev);
1006 	if (err != 0) {
1007 		pr_err("Could not register Xen privcmd device\n");
1008 		return err;
1009 	}
1010 
1011 	err = misc_register(&xen_privcmdbuf_dev);
1012 	if (err != 0) {
1013 		pr_err("Could not register Xen hypercall-buf device\n");
1014 		misc_deregister(&privcmd_dev);
1015 		return err;
1016 	}
1017 
1018 	return 0;
1019 }
1020 
1021 static void __exit privcmd_exit(void)
1022 {
1023 	misc_deregister(&privcmd_dev);
1024 	misc_deregister(&xen_privcmdbuf_dev);
1025 }
1026 
1027 module_init(privcmd_init);
1028 module_exit(privcmd_exit);
1029