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