xref: /openbmc/linux/drivers/xen/privcmd.c (revision 9b93eb47)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /******************************************************************************
3  * privcmd.c
4  *
5  * Interface to privileged domain-0 commands.
6  *
7  * Copyright (c) 2002-2004, K A Fraser, B Dragovic
8  */
9 
10 #define pr_fmt(fmt) "xen:" KBUILD_MODNAME ": " fmt
11 
12 #include <linux/kernel.h>
13 #include <linux/module.h>
14 #include <linux/sched.h>
15 #include <linux/slab.h>
16 #include <linux/string.h>
17 #include <linux/errno.h>
18 #include <linux/mm.h>
19 #include <linux/mman.h>
20 #include <linux/uaccess.h>
21 #include <linux/swap.h>
22 #include <linux/highmem.h>
23 #include <linux/pagemap.h>
24 #include <linux/seq_file.h>
25 #include <linux/miscdevice.h>
26 #include <linux/moduleparam.h>
27 
28 #include <asm/pgalloc.h>
29 #include <asm/pgtable.h>
30 #include <asm/tlb.h>
31 #include <asm/xen/hypervisor.h>
32 #include <asm/xen/hypercall.h>
33 
34 #include <xen/xen.h>
35 #include <xen/privcmd.h>
36 #include <xen/interface/xen.h>
37 #include <xen/interface/memory.h>
38 #include <xen/interface/hvm/dm_op.h>
39 #include <xen/features.h>
40 #include <xen/page.h>
41 #include <xen/xen-ops.h>
42 #include <xen/balloon.h>
43 
44 #include "privcmd.h"
45 
46 MODULE_LICENSE("GPL");
47 
48 #define PRIV_VMA_LOCKED ((void *)1)
49 
50 static unsigned int privcmd_dm_op_max_num = 16;
51 module_param_named(dm_op_max_nr_bufs, privcmd_dm_op_max_num, uint, 0644);
52 MODULE_PARM_DESC(dm_op_max_nr_bufs,
53 		 "Maximum number of buffers per dm_op hypercall");
54 
55 static unsigned int privcmd_dm_op_buf_max_size = 4096;
56 module_param_named(dm_op_buf_max_size, privcmd_dm_op_buf_max_size, uint,
57 		   0644);
58 MODULE_PARM_DESC(dm_op_buf_max_size,
59 		 "Maximum size of a dm_op hypercall buffer");
60 
61 struct privcmd_data {
62 	domid_t domid;
63 };
64 
65 static int privcmd_vma_range_is_mapped(
66                struct vm_area_struct *vma,
67                unsigned long addr,
68                unsigned long nr_pages);
69 
70 static long privcmd_ioctl_hypercall(struct file *file, void __user *udata)
71 {
72 	struct privcmd_data *data = file->private_data;
73 	struct privcmd_hypercall hypercall;
74 	long ret;
75 
76 	/* Disallow arbitrary hypercalls if restricted */
77 	if (data->domid != DOMID_INVALID)
78 		return -EPERM;
79 
80 	if (copy_from_user(&hypercall, udata, sizeof(hypercall)))
81 		return -EFAULT;
82 
83 	xen_preemptible_hcall_begin();
84 	ret = privcmd_call(hypercall.op,
85 			   hypercall.arg[0], hypercall.arg[1],
86 			   hypercall.arg[2], hypercall.arg[3],
87 			   hypercall.arg[4]);
88 	xen_preemptible_hcall_end();
89 
90 	return ret;
91 }
92 
93 static void free_page_list(struct list_head *pages)
94 {
95 	struct page *p, *n;
96 
97 	list_for_each_entry_safe(p, n, pages, lru)
98 		__free_page(p);
99 
100 	INIT_LIST_HEAD(pages);
101 }
102 
103 /*
104  * Given an array of items in userspace, return a list of pages
105  * containing the data.  If copying fails, either because of memory
106  * allocation failure or a problem reading user memory, return an
107  * error code; its up to the caller to dispose of any partial list.
108  */
109 static int gather_array(struct list_head *pagelist,
110 			unsigned nelem, size_t size,
111 			const void __user *data)
112 {
113 	unsigned pageidx;
114 	void *pagedata;
115 	int ret;
116 
117 	if (size > PAGE_SIZE)
118 		return 0;
119 
120 	pageidx = PAGE_SIZE;
121 	pagedata = NULL;	/* quiet, gcc */
122 	while (nelem--) {
123 		if (pageidx > PAGE_SIZE-size) {
124 			struct page *page = alloc_page(GFP_KERNEL);
125 
126 			ret = -ENOMEM;
127 			if (page == NULL)
128 				goto fail;
129 
130 			pagedata = page_address(page);
131 
132 			list_add_tail(&page->lru, pagelist);
133 			pageidx = 0;
134 		}
135 
136 		ret = -EFAULT;
137 		if (copy_from_user(pagedata + pageidx, data, size))
138 			goto fail;
139 
140 		data += size;
141 		pageidx += size;
142 	}
143 
144 	ret = 0;
145 
146 fail:
147 	return ret;
148 }
149 
150 /*
151  * Call function "fn" on each element of the array fragmented
152  * over a list of pages.
153  */
154 static int traverse_pages(unsigned nelem, size_t size,
155 			  struct list_head *pos,
156 			  int (*fn)(void *data, void *state),
157 			  void *state)
158 {
159 	void *pagedata;
160 	unsigned pageidx;
161 	int ret = 0;
162 
163 	BUG_ON(size > PAGE_SIZE);
164 
165 	pageidx = PAGE_SIZE;
166 	pagedata = NULL;	/* hush, gcc */
167 
168 	while (nelem--) {
169 		if (pageidx > PAGE_SIZE-size) {
170 			struct page *page;
171 			pos = pos->next;
172 			page = list_entry(pos, struct page, lru);
173 			pagedata = page_address(page);
174 			pageidx = 0;
175 		}
176 
177 		ret = (*fn)(pagedata + pageidx, state);
178 		if (ret)
179 			break;
180 		pageidx += size;
181 	}
182 
183 	return ret;
184 }
185 
186 /*
187  * Similar to traverse_pages, but use each page as a "block" of
188  * data to be processed as one unit.
189  */
190 static int traverse_pages_block(unsigned nelem, size_t size,
191 				struct list_head *pos,
192 				int (*fn)(void *data, int nr, void *state),
193 				void *state)
194 {
195 	void *pagedata;
196 	int ret = 0;
197 
198 	BUG_ON(size > PAGE_SIZE);
199 
200 	while (nelem) {
201 		int nr = (PAGE_SIZE/size);
202 		struct page *page;
203 		if (nr > nelem)
204 			nr = nelem;
205 		pos = pos->next;
206 		page = list_entry(pos, struct page, lru);
207 		pagedata = page_address(page);
208 		ret = (*fn)(pagedata, nr, state);
209 		if (ret)
210 			break;
211 		nelem -= nr;
212 	}
213 
214 	return ret;
215 }
216 
217 struct mmap_gfn_state {
218 	unsigned long va;
219 	struct vm_area_struct *vma;
220 	domid_t domain;
221 };
222 
223 static int mmap_gfn_range(void *data, void *state)
224 {
225 	struct privcmd_mmap_entry *msg = data;
226 	struct mmap_gfn_state *st = state;
227 	struct vm_area_struct *vma = st->vma;
228 	int rc;
229 
230 	/* Do not allow range to wrap the address space. */
231 	if ((msg->npages > (LONG_MAX >> PAGE_SHIFT)) ||
232 	    ((unsigned long)(msg->npages << PAGE_SHIFT) >= -st->va))
233 		return -EINVAL;
234 
235 	/* Range chunks must be contiguous in va space. */
236 	if ((msg->va != st->va) ||
237 	    ((msg->va+(msg->npages<<PAGE_SHIFT)) > vma->vm_end))
238 		return -EINVAL;
239 
240 	rc = xen_remap_domain_gfn_range(vma,
241 					msg->va & PAGE_MASK,
242 					msg->mfn, msg->npages,
243 					vma->vm_page_prot,
244 					st->domain, NULL);
245 	if (rc < 0)
246 		return rc;
247 
248 	st->va += msg->npages << PAGE_SHIFT;
249 
250 	return 0;
251 }
252 
253 static long privcmd_ioctl_mmap(struct file *file, void __user *udata)
254 {
255 	struct privcmd_data *data = file->private_data;
256 	struct privcmd_mmap mmapcmd;
257 	struct mm_struct *mm = current->mm;
258 	struct vm_area_struct *vma;
259 	int rc;
260 	LIST_HEAD(pagelist);
261 	struct mmap_gfn_state state;
262 
263 	/* We only support privcmd_ioctl_mmap_batch for auto translated. */
264 	if (xen_feature(XENFEAT_auto_translated_physmap))
265 		return -ENOSYS;
266 
267 	if (copy_from_user(&mmapcmd, udata, sizeof(mmapcmd)))
268 		return -EFAULT;
269 
270 	/* If restriction is in place, check the domid matches */
271 	if (data->domid != DOMID_INVALID && data->domid != mmapcmd.dom)
272 		return -EPERM;
273 
274 	rc = gather_array(&pagelist,
275 			  mmapcmd.num, sizeof(struct privcmd_mmap_entry),
276 			  mmapcmd.entry);
277 
278 	if (rc || list_empty(&pagelist))
279 		goto out;
280 
281 	down_write(&mm->mmap_sem);
282 
283 	{
284 		struct page *page = list_first_entry(&pagelist,
285 						     struct page, lru);
286 		struct privcmd_mmap_entry *msg = page_address(page);
287 
288 		vma = find_vma(mm, msg->va);
289 		rc = -EINVAL;
290 
291 		if (!vma || (msg->va != vma->vm_start) || vma->vm_private_data)
292 			goto out_up;
293 		vma->vm_private_data = PRIV_VMA_LOCKED;
294 	}
295 
296 	state.va = vma->vm_start;
297 	state.vma = vma;
298 	state.domain = mmapcmd.dom;
299 
300 	rc = traverse_pages(mmapcmd.num, sizeof(struct privcmd_mmap_entry),
301 			    &pagelist,
302 			    mmap_gfn_range, &state);
303 
304 
305 out_up:
306 	up_write(&mm->mmap_sem);
307 
308 out:
309 	free_page_list(&pagelist);
310 
311 	return rc;
312 }
313 
314 struct mmap_batch_state {
315 	domid_t domain;
316 	unsigned long va;
317 	struct vm_area_struct *vma;
318 	int index;
319 	/* A tristate:
320 	 *      0 for no errors
321 	 *      1 if at least one error has happened (and no
322 	 *          -ENOENT errors have happened)
323 	 *      -ENOENT if at least 1 -ENOENT has happened.
324 	 */
325 	int global_error;
326 	int version;
327 
328 	/* User-space gfn array to store errors in the second pass for V1. */
329 	xen_pfn_t __user *user_gfn;
330 	/* User-space int array to store errors in the second pass for V2. */
331 	int __user *user_err;
332 };
333 
334 /* auto translated dom0 note: if domU being created is PV, then gfn is
335  * mfn(addr on bus). If it's auto xlated, then gfn is pfn (input to HAP).
336  */
337 static int mmap_batch_fn(void *data, int nr, void *state)
338 {
339 	xen_pfn_t *gfnp = data;
340 	struct mmap_batch_state *st = state;
341 	struct vm_area_struct *vma = st->vma;
342 	struct page **pages = vma->vm_private_data;
343 	struct page **cur_pages = NULL;
344 	int ret;
345 
346 	if (xen_feature(XENFEAT_auto_translated_physmap))
347 		cur_pages = &pages[st->index];
348 
349 	BUG_ON(nr < 0);
350 	ret = xen_remap_domain_gfn_array(st->vma, st->va & PAGE_MASK, gfnp, nr,
351 					 (int *)gfnp, st->vma->vm_page_prot,
352 					 st->domain, cur_pages);
353 
354 	/* Adjust the global_error? */
355 	if (ret != nr) {
356 		if (ret == -ENOENT)
357 			st->global_error = -ENOENT;
358 		else {
359 			/* Record that at least one error has happened. */
360 			if (st->global_error == 0)
361 				st->global_error = 1;
362 		}
363 	}
364 	st->va += XEN_PAGE_SIZE * nr;
365 	st->index += nr / XEN_PFN_PER_PAGE;
366 
367 	return 0;
368 }
369 
370 static int mmap_return_error(int err, struct mmap_batch_state *st)
371 {
372 	int ret;
373 
374 	if (st->version == 1) {
375 		if (err) {
376 			xen_pfn_t gfn;
377 
378 			ret = get_user(gfn, st->user_gfn);
379 			if (ret < 0)
380 				return ret;
381 			/*
382 			 * V1 encodes the error codes in the 32bit top
383 			 * nibble of the gfn (with its known
384 			 * limitations vis-a-vis 64 bit callers).
385 			 */
386 			gfn |= (err == -ENOENT) ?
387 				PRIVCMD_MMAPBATCH_PAGED_ERROR :
388 				PRIVCMD_MMAPBATCH_MFN_ERROR;
389 			return __put_user(gfn, st->user_gfn++);
390 		} else
391 			st->user_gfn++;
392 	} else { /* st->version == 2 */
393 		if (err)
394 			return __put_user(err, st->user_err++);
395 		else
396 			st->user_err++;
397 	}
398 
399 	return 0;
400 }
401 
402 static int mmap_return_errors(void *data, int nr, void *state)
403 {
404 	struct mmap_batch_state *st = state;
405 	int *errs = data;
406 	int i;
407 	int ret;
408 
409 	for (i = 0; i < nr; i++) {
410 		ret = mmap_return_error(errs[i], st);
411 		if (ret < 0)
412 			return ret;
413 	}
414 	return 0;
415 }
416 
417 /* Allocate pfns that are then mapped with gfns from foreign domid. Update
418  * the vma with the page info to use later.
419  * Returns: 0 if success, otherwise -errno
420  */
421 static int alloc_empty_pages(struct vm_area_struct *vma, int numpgs)
422 {
423 	int rc;
424 	struct page **pages;
425 
426 	pages = kcalloc(numpgs, sizeof(pages[0]), GFP_KERNEL);
427 	if (pages == NULL)
428 		return -ENOMEM;
429 
430 	rc = alloc_xenballooned_pages(numpgs, pages);
431 	if (rc != 0) {
432 		pr_warn("%s Could not alloc %d pfns rc:%d\n", __func__,
433 			numpgs, rc);
434 		kfree(pages);
435 		return -ENOMEM;
436 	}
437 	BUG_ON(vma->vm_private_data != NULL);
438 	vma->vm_private_data = pages;
439 
440 	return 0;
441 }
442 
443 static const struct vm_operations_struct privcmd_vm_ops;
444 
445 static long privcmd_ioctl_mmap_batch(
446 	struct file *file, void __user *udata, int version)
447 {
448 	struct privcmd_data *data = file->private_data;
449 	int ret;
450 	struct privcmd_mmapbatch_v2 m;
451 	struct mm_struct *mm = current->mm;
452 	struct vm_area_struct *vma;
453 	unsigned long nr_pages;
454 	LIST_HEAD(pagelist);
455 	struct mmap_batch_state state;
456 
457 	switch (version) {
458 	case 1:
459 		if (copy_from_user(&m, udata, sizeof(struct privcmd_mmapbatch)))
460 			return -EFAULT;
461 		/* Returns per-frame error in m.arr. */
462 		m.err = NULL;
463 		if (!access_ok(m.arr, m.num * sizeof(*m.arr)))
464 			return -EFAULT;
465 		break;
466 	case 2:
467 		if (copy_from_user(&m, udata, sizeof(struct privcmd_mmapbatch_v2)))
468 			return -EFAULT;
469 		/* Returns per-frame error code in m.err. */
470 		if (!access_ok(m.err, m.num * (sizeof(*m.err))))
471 			return -EFAULT;
472 		break;
473 	default:
474 		return -EINVAL;
475 	}
476 
477 	/* If restriction is in place, check the domid matches */
478 	if (data->domid != DOMID_INVALID && data->domid != m.dom)
479 		return -EPERM;
480 
481 	nr_pages = DIV_ROUND_UP(m.num, XEN_PFN_PER_PAGE);
482 	if ((m.num <= 0) || (nr_pages > (LONG_MAX >> PAGE_SHIFT)))
483 		return -EINVAL;
484 
485 	ret = gather_array(&pagelist, m.num, sizeof(xen_pfn_t), m.arr);
486 
487 	if (ret)
488 		goto out;
489 	if (list_empty(&pagelist)) {
490 		ret = -EINVAL;
491 		goto out;
492 	}
493 
494 	if (version == 2) {
495 		/* Zero error array now to only copy back actual errors. */
496 		if (clear_user(m.err, sizeof(int) * m.num)) {
497 			ret = -EFAULT;
498 			goto out;
499 		}
500 	}
501 
502 	down_write(&mm->mmap_sem);
503 
504 	vma = find_vma(mm, m.addr);
505 	if (!vma ||
506 	    vma->vm_ops != &privcmd_vm_ops) {
507 		ret = -EINVAL;
508 		goto out_unlock;
509 	}
510 
511 	/*
512 	 * Caller must either:
513 	 *
514 	 * Map the whole VMA range, which will also allocate all the
515 	 * pages required for the auto_translated_physmap case.
516 	 *
517 	 * Or
518 	 *
519 	 * Map unmapped holes left from a previous map attempt (e.g.,
520 	 * because those foreign frames were previously paged out).
521 	 */
522 	if (vma->vm_private_data == NULL) {
523 		if (m.addr != vma->vm_start ||
524 		    m.addr + (nr_pages << PAGE_SHIFT) != vma->vm_end) {
525 			ret = -EINVAL;
526 			goto out_unlock;
527 		}
528 		if (xen_feature(XENFEAT_auto_translated_physmap)) {
529 			ret = alloc_empty_pages(vma, nr_pages);
530 			if (ret < 0)
531 				goto out_unlock;
532 		} else
533 			vma->vm_private_data = PRIV_VMA_LOCKED;
534 	} else {
535 		if (m.addr < vma->vm_start ||
536 		    m.addr + (nr_pages << PAGE_SHIFT) > vma->vm_end) {
537 			ret = -EINVAL;
538 			goto out_unlock;
539 		}
540 		if (privcmd_vma_range_is_mapped(vma, m.addr, nr_pages)) {
541 			ret = -EINVAL;
542 			goto out_unlock;
543 		}
544 	}
545 
546 	state.domain        = m.dom;
547 	state.vma           = vma;
548 	state.va            = m.addr;
549 	state.index         = 0;
550 	state.global_error  = 0;
551 	state.version       = version;
552 
553 	BUILD_BUG_ON(((PAGE_SIZE / sizeof(xen_pfn_t)) % XEN_PFN_PER_PAGE) != 0);
554 	/* mmap_batch_fn guarantees ret == 0 */
555 	BUG_ON(traverse_pages_block(m.num, sizeof(xen_pfn_t),
556 				    &pagelist, mmap_batch_fn, &state));
557 
558 	up_write(&mm->mmap_sem);
559 
560 	if (state.global_error) {
561 		/* Write back errors in second pass. */
562 		state.user_gfn = (xen_pfn_t *)m.arr;
563 		state.user_err = m.err;
564 		ret = traverse_pages_block(m.num, sizeof(xen_pfn_t),
565 					   &pagelist, mmap_return_errors, &state);
566 	} else
567 		ret = 0;
568 
569 	/* If we have not had any EFAULT-like global errors then set the global
570 	 * error to -ENOENT if necessary. */
571 	if ((ret == 0) && (state.global_error == -ENOENT))
572 		ret = -ENOENT;
573 
574 out:
575 	free_page_list(&pagelist);
576 	return ret;
577 
578 out_unlock:
579 	up_write(&mm->mmap_sem);
580 	goto out;
581 }
582 
583 static int lock_pages(
584 	struct privcmd_dm_op_buf kbufs[], unsigned int num,
585 	struct page *pages[], unsigned int nr_pages)
586 {
587 	unsigned int i;
588 
589 	for (i = 0; i < num; i++) {
590 		unsigned int requested;
591 		int pinned;
592 
593 		requested = DIV_ROUND_UP(
594 			offset_in_page(kbufs[i].uptr) + kbufs[i].size,
595 			PAGE_SIZE);
596 		if (requested > nr_pages)
597 			return -ENOSPC;
598 
599 		pinned = get_user_pages_fast(
600 			(unsigned long) kbufs[i].uptr,
601 			requested, FOLL_WRITE, pages);
602 		if (pinned < 0)
603 			return pinned;
604 
605 		nr_pages -= pinned;
606 		pages += pinned;
607 	}
608 
609 	return 0;
610 }
611 
612 static void unlock_pages(struct page *pages[], unsigned int nr_pages)
613 {
614 	unsigned int i;
615 
616 	if (!pages)
617 		return;
618 
619 	for (i = 0; i < nr_pages; i++) {
620 		if (pages[i])
621 			put_page(pages[i]);
622 	}
623 }
624 
625 static long privcmd_ioctl_dm_op(struct file *file, void __user *udata)
626 {
627 	struct privcmd_data *data = file->private_data;
628 	struct privcmd_dm_op kdata;
629 	struct privcmd_dm_op_buf *kbufs;
630 	unsigned int nr_pages = 0;
631 	struct page **pages = NULL;
632 	struct xen_dm_op_buf *xbufs = NULL;
633 	unsigned int i;
634 	long rc;
635 
636 	if (copy_from_user(&kdata, udata, sizeof(kdata)))
637 		return -EFAULT;
638 
639 	/* If restriction is in place, check the domid matches */
640 	if (data->domid != DOMID_INVALID && data->domid != kdata.dom)
641 		return -EPERM;
642 
643 	if (kdata.num == 0)
644 		return 0;
645 
646 	if (kdata.num > privcmd_dm_op_max_num)
647 		return -E2BIG;
648 
649 	kbufs = kcalloc(kdata.num, sizeof(*kbufs), GFP_KERNEL);
650 	if (!kbufs)
651 		return -ENOMEM;
652 
653 	if (copy_from_user(kbufs, kdata.ubufs,
654 			   sizeof(*kbufs) * kdata.num)) {
655 		rc = -EFAULT;
656 		goto out;
657 	}
658 
659 	for (i = 0; i < kdata.num; i++) {
660 		if (kbufs[i].size > privcmd_dm_op_buf_max_size) {
661 			rc = -E2BIG;
662 			goto out;
663 		}
664 
665 		if (!access_ok(kbufs[i].uptr,
666 			       kbufs[i].size)) {
667 			rc = -EFAULT;
668 			goto out;
669 		}
670 
671 		nr_pages += DIV_ROUND_UP(
672 			offset_in_page(kbufs[i].uptr) + kbufs[i].size,
673 			PAGE_SIZE);
674 	}
675 
676 	pages = kcalloc(nr_pages, sizeof(*pages), GFP_KERNEL);
677 	if (!pages) {
678 		rc = -ENOMEM;
679 		goto out;
680 	}
681 
682 	xbufs = kcalloc(kdata.num, sizeof(*xbufs), GFP_KERNEL);
683 	if (!xbufs) {
684 		rc = -ENOMEM;
685 		goto out;
686 	}
687 
688 	rc = lock_pages(kbufs, kdata.num, pages, nr_pages);
689 	if (rc)
690 		goto out;
691 
692 	for (i = 0; i < kdata.num; i++) {
693 		set_xen_guest_handle(xbufs[i].h, kbufs[i].uptr);
694 		xbufs[i].size = kbufs[i].size;
695 	}
696 
697 	xen_preemptible_hcall_begin();
698 	rc = HYPERVISOR_dm_op(kdata.dom, kdata.num, xbufs);
699 	xen_preemptible_hcall_end();
700 
701 out:
702 	unlock_pages(pages, nr_pages);
703 	kfree(xbufs);
704 	kfree(pages);
705 	kfree(kbufs);
706 
707 	return rc;
708 }
709 
710 static long privcmd_ioctl_restrict(struct file *file, void __user *udata)
711 {
712 	struct privcmd_data *data = file->private_data;
713 	domid_t dom;
714 
715 	if (copy_from_user(&dom, udata, sizeof(dom)))
716 		return -EFAULT;
717 
718 	/* Set restriction to the specified domain, or check it matches */
719 	if (data->domid == DOMID_INVALID)
720 		data->domid = dom;
721 	else if (data->domid != dom)
722 		return -EINVAL;
723 
724 	return 0;
725 }
726 
727 struct remap_pfn {
728 	struct mm_struct *mm;
729 	struct page **pages;
730 	pgprot_t prot;
731 	unsigned long i;
732 };
733 
734 static int remap_pfn_fn(pte_t *ptep, pgtable_t token, unsigned long addr,
735 			void *data)
736 {
737 	struct remap_pfn *r = data;
738 	struct page *page = r->pages[r->i];
739 	pte_t pte = pte_mkspecial(pfn_pte(page_to_pfn(page), r->prot));
740 
741 	set_pte_at(r->mm, addr, ptep, pte);
742 	r->i++;
743 
744 	return 0;
745 }
746 
747 static long privcmd_ioctl_mmap_resource(struct file *file, void __user *udata)
748 {
749 	struct privcmd_data *data = file->private_data;
750 	struct mm_struct *mm = current->mm;
751 	struct vm_area_struct *vma;
752 	struct privcmd_mmap_resource kdata;
753 	xen_pfn_t *pfns = NULL;
754 	struct xen_mem_acquire_resource xdata;
755 	int rc;
756 
757 	if (copy_from_user(&kdata, udata, sizeof(kdata)))
758 		return -EFAULT;
759 
760 	/* If restriction is in place, check the domid matches */
761 	if (data->domid != DOMID_INVALID && data->domid != kdata.dom)
762 		return -EPERM;
763 
764 	down_write(&mm->mmap_sem);
765 
766 	vma = find_vma(mm, kdata.addr);
767 	if (!vma || vma->vm_ops != &privcmd_vm_ops) {
768 		rc = -EINVAL;
769 		goto out;
770 	}
771 
772 	pfns = kcalloc(kdata.num, sizeof(*pfns), GFP_KERNEL);
773 	if (!pfns) {
774 		rc = -ENOMEM;
775 		goto out;
776 	}
777 
778 	if (xen_feature(XENFEAT_auto_translated_physmap)) {
779 		unsigned int nr = DIV_ROUND_UP(kdata.num, XEN_PFN_PER_PAGE);
780 		struct page **pages;
781 		unsigned int i;
782 
783 		rc = alloc_empty_pages(vma, nr);
784 		if (rc < 0)
785 			goto out;
786 
787 		pages = vma->vm_private_data;
788 		for (i = 0; i < kdata.num; i++) {
789 			xen_pfn_t pfn =
790 				page_to_xen_pfn(pages[i / XEN_PFN_PER_PAGE]);
791 
792 			pfns[i] = pfn + (i % XEN_PFN_PER_PAGE);
793 		}
794 	} else
795 		vma->vm_private_data = PRIV_VMA_LOCKED;
796 
797 	memset(&xdata, 0, sizeof(xdata));
798 	xdata.domid = kdata.dom;
799 	xdata.type = kdata.type;
800 	xdata.id = kdata.id;
801 	xdata.frame = kdata.idx;
802 	xdata.nr_frames = kdata.num;
803 	set_xen_guest_handle(xdata.frame_list, pfns);
804 
805 	xen_preemptible_hcall_begin();
806 	rc = HYPERVISOR_memory_op(XENMEM_acquire_resource, &xdata);
807 	xen_preemptible_hcall_end();
808 
809 	if (rc)
810 		goto out;
811 
812 	if (xen_feature(XENFEAT_auto_translated_physmap)) {
813 		struct remap_pfn r = {
814 			.mm = vma->vm_mm,
815 			.pages = vma->vm_private_data,
816 			.prot = vma->vm_page_prot,
817 		};
818 
819 		rc = apply_to_page_range(r.mm, kdata.addr,
820 					 kdata.num << PAGE_SHIFT,
821 					 remap_pfn_fn, &r);
822 	} else {
823 		unsigned int domid =
824 			(xdata.flags & XENMEM_rsrc_acq_caller_owned) ?
825 			DOMID_SELF : kdata.dom;
826 		int num;
827 
828 		num = xen_remap_domain_mfn_array(vma,
829 						 kdata.addr & PAGE_MASK,
830 						 pfns, kdata.num, (int *)pfns,
831 						 vma->vm_page_prot,
832 						 domid,
833 						 vma->vm_private_data);
834 		if (num < 0)
835 			rc = num;
836 		else if (num != kdata.num) {
837 			unsigned int i;
838 
839 			for (i = 0; i < num; i++) {
840 				rc = pfns[i];
841 				if (rc < 0)
842 					break;
843 			}
844 		} else
845 			rc = 0;
846 	}
847 
848 out:
849 	up_write(&mm->mmap_sem);
850 	kfree(pfns);
851 
852 	return rc;
853 }
854 
855 static long privcmd_ioctl(struct file *file,
856 			  unsigned int cmd, unsigned long data)
857 {
858 	int ret = -ENOTTY;
859 	void __user *udata = (void __user *) data;
860 
861 	switch (cmd) {
862 	case IOCTL_PRIVCMD_HYPERCALL:
863 		ret = privcmd_ioctl_hypercall(file, udata);
864 		break;
865 
866 	case IOCTL_PRIVCMD_MMAP:
867 		ret = privcmd_ioctl_mmap(file, udata);
868 		break;
869 
870 	case IOCTL_PRIVCMD_MMAPBATCH:
871 		ret = privcmd_ioctl_mmap_batch(file, udata, 1);
872 		break;
873 
874 	case IOCTL_PRIVCMD_MMAPBATCH_V2:
875 		ret = privcmd_ioctl_mmap_batch(file, udata, 2);
876 		break;
877 
878 	case IOCTL_PRIVCMD_DM_OP:
879 		ret = privcmd_ioctl_dm_op(file, udata);
880 		break;
881 
882 	case IOCTL_PRIVCMD_RESTRICT:
883 		ret = privcmd_ioctl_restrict(file, udata);
884 		break;
885 
886 	case IOCTL_PRIVCMD_MMAP_RESOURCE:
887 		ret = privcmd_ioctl_mmap_resource(file, udata);
888 		break;
889 
890 	default:
891 		break;
892 	}
893 
894 	return ret;
895 }
896 
897 static int privcmd_open(struct inode *ino, struct file *file)
898 {
899 	struct privcmd_data *data = kzalloc(sizeof(*data), GFP_KERNEL);
900 
901 	if (!data)
902 		return -ENOMEM;
903 
904 	/* DOMID_INVALID implies no restriction */
905 	data->domid = DOMID_INVALID;
906 
907 	file->private_data = data;
908 	return 0;
909 }
910 
911 static int privcmd_release(struct inode *ino, struct file *file)
912 {
913 	struct privcmd_data *data = file->private_data;
914 
915 	kfree(data);
916 	return 0;
917 }
918 
919 static void privcmd_close(struct vm_area_struct *vma)
920 {
921 	struct page **pages = vma->vm_private_data;
922 	int numpgs = vma_pages(vma);
923 	int numgfns = (vma->vm_end - vma->vm_start) >> XEN_PAGE_SHIFT;
924 	int rc;
925 
926 	if (!xen_feature(XENFEAT_auto_translated_physmap) || !numpgs || !pages)
927 		return;
928 
929 	rc = xen_unmap_domain_gfn_range(vma, numgfns, pages);
930 	if (rc == 0)
931 		free_xenballooned_pages(numpgs, pages);
932 	else
933 		pr_crit("unable to unmap MFN range: leaking %d pages. rc=%d\n",
934 			numpgs, rc);
935 	kfree(pages);
936 }
937 
938 static vm_fault_t privcmd_fault(struct vm_fault *vmf)
939 {
940 	printk(KERN_DEBUG "privcmd_fault: vma=%p %lx-%lx, pgoff=%lx, uv=%p\n",
941 	       vmf->vma, vmf->vma->vm_start, vmf->vma->vm_end,
942 	       vmf->pgoff, (void *)vmf->address);
943 
944 	return VM_FAULT_SIGBUS;
945 }
946 
947 static const struct vm_operations_struct privcmd_vm_ops = {
948 	.close = privcmd_close,
949 	.fault = privcmd_fault
950 };
951 
952 static int privcmd_mmap(struct file *file, struct vm_area_struct *vma)
953 {
954 	/* DONTCOPY is essential for Xen because copy_page_range doesn't know
955 	 * how to recreate these mappings */
956 	vma->vm_flags |= VM_IO | VM_PFNMAP | VM_DONTCOPY |
957 			 VM_DONTEXPAND | VM_DONTDUMP;
958 	vma->vm_ops = &privcmd_vm_ops;
959 	vma->vm_private_data = NULL;
960 
961 	return 0;
962 }
963 
964 /*
965  * For MMAPBATCH*. This allows asserting the singleshot mapping
966  * on a per pfn/pte basis. Mapping calls that fail with ENOENT
967  * can be then retried until success.
968  */
969 static int is_mapped_fn(pte_t *pte, struct page *pmd_page,
970 	                unsigned long addr, void *data)
971 {
972 	return pte_none(*pte) ? 0 : -EBUSY;
973 }
974 
975 static int privcmd_vma_range_is_mapped(
976 	           struct vm_area_struct *vma,
977 	           unsigned long addr,
978 	           unsigned long nr_pages)
979 {
980 	return apply_to_page_range(vma->vm_mm, addr, nr_pages << PAGE_SHIFT,
981 				   is_mapped_fn, NULL) != 0;
982 }
983 
984 const struct file_operations xen_privcmd_fops = {
985 	.owner = THIS_MODULE,
986 	.unlocked_ioctl = privcmd_ioctl,
987 	.open = privcmd_open,
988 	.release = privcmd_release,
989 	.mmap = privcmd_mmap,
990 };
991 EXPORT_SYMBOL_GPL(xen_privcmd_fops);
992 
993 static struct miscdevice privcmd_dev = {
994 	.minor = MISC_DYNAMIC_MINOR,
995 	.name = "xen/privcmd",
996 	.fops = &xen_privcmd_fops,
997 };
998 
999 static int __init privcmd_init(void)
1000 {
1001 	int err;
1002 
1003 	if (!xen_domain())
1004 		return -ENODEV;
1005 
1006 	err = misc_register(&privcmd_dev);
1007 	if (err != 0) {
1008 		pr_err("Could not register Xen privcmd device\n");
1009 		return err;
1010 	}
1011 
1012 	err = misc_register(&xen_privcmdbuf_dev);
1013 	if (err != 0) {
1014 		pr_err("Could not register Xen hypercall-buf device\n");
1015 		misc_deregister(&privcmd_dev);
1016 		return err;
1017 	}
1018 
1019 	return 0;
1020 }
1021 
1022 static void __exit privcmd_exit(void)
1023 {
1024 	misc_deregister(&privcmd_dev);
1025 	misc_deregister(&xen_privcmdbuf_dev);
1026 }
1027 
1028 module_init(privcmd_init);
1029 module_exit(privcmd_exit);
1030