1 /******************************************************************************
4 * Interface to privileged domain-0 commands.
6 * Copyright (c) 2002-2004, K A Fraser, B Dragovic
9 #define pr_fmt(fmt) "xen:" KBUILD_MODNAME ": " fmt
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>
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>
27 #include <asm/pgalloc.h>
28 #include <asm/pgtable.h>
30 #include <asm/xen/hypervisor.h>
31 #include <asm/xen/hypercall.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>
40 #include <xen/xen-ops.h>
41 #include <xen/balloon.h>
45 MODULE_LICENSE("GPL");
47 #define PRIV_VMA_LOCKED ((void *)1)
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");
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,
57 MODULE_PARM_DESC(dm_op_buf_max_size,
58 "Maximum size of a dm_op hypercall buffer");
64 static int privcmd_vma_range_is_mapped(
65 struct vm_area_struct *vma,
67 unsigned long nr_pages);
69 static long privcmd_ioctl_hypercall(struct file *file, void __user *udata)
71 struct privcmd_data *data = file->private_data;
72 struct privcmd_hypercall hypercall;
75 /* Disallow arbitrary hypercalls if restricted */
76 if (data->domid != DOMID_INVALID)
79 if (copy_from_user(&hypercall, udata, sizeof(hypercall)))
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],
87 xen_preemptible_hcall_end();
92 static void free_page_list(struct list_head *pages)
96 list_for_each_entry_safe(p, n, pages, lru)
99 INIT_LIST_HEAD(pages);
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.
108 static int gather_array(struct list_head *pagelist,
109 unsigned nelem, size_t size,
110 const void __user *data)
116 if (size > PAGE_SIZE)
120 pagedata = NULL; /* quiet, gcc */
122 if (pageidx > PAGE_SIZE-size) {
123 struct page *page = alloc_page(GFP_KERNEL);
129 pagedata = page_address(page);
131 list_add_tail(&page->lru, pagelist);
136 if (copy_from_user(pagedata + pageidx, data, size))
150 * Call function "fn" on each element of the array fragmented
151 * over a list of pages.
153 static int traverse_pages(unsigned nelem, size_t size,
154 struct list_head *pos,
155 int (*fn)(void *data, void *state),
162 BUG_ON(size > PAGE_SIZE);
165 pagedata = NULL; /* hush, gcc */
168 if (pageidx > PAGE_SIZE-size) {
171 page = list_entry(pos, struct page, lru);
172 pagedata = page_address(page);
176 ret = (*fn)(pagedata + pageidx, state);
186 * Similar to traverse_pages, but use each page as a "block" of
187 * data to be processed as one unit.
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),
197 BUG_ON(size > PAGE_SIZE);
200 int nr = (PAGE_SIZE/size);
205 page = list_entry(pos, struct page, lru);
206 pagedata = page_address(page);
207 ret = (*fn)(pagedata, nr, state);
216 struct mmap_gfn_state {
218 struct vm_area_struct *vma;
222 static int mmap_gfn_range(void *data, void *state)
224 struct privcmd_mmap_entry *msg = data;
225 struct mmap_gfn_state *st = state;
226 struct vm_area_struct *vma = st->vma;
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))
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))
239 rc = xen_remap_domain_gfn_range(vma,
241 msg->mfn, msg->npages,
247 st->va += msg->npages << PAGE_SHIFT;
252 static long privcmd_ioctl_mmap(struct file *file, void __user *udata)
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;
260 struct mmap_gfn_state state;
262 /* We only support privcmd_ioctl_mmap_batch for auto translated. */
263 if (xen_feature(XENFEAT_auto_translated_physmap))
266 if (copy_from_user(&mmapcmd, udata, sizeof(mmapcmd)))
269 /* If restriction is in place, check the domid matches */
270 if (data->domid != DOMID_INVALID && data->domid != mmapcmd.dom)
273 rc = gather_array(&pagelist,
274 mmapcmd.num, sizeof(struct privcmd_mmap_entry),
277 if (rc || list_empty(&pagelist))
280 down_write(&mm->mmap_sem);
283 struct page *page = list_first_entry(&pagelist,
285 struct privcmd_mmap_entry *msg = page_address(page);
287 vma = find_vma(mm, msg->va);
290 if (!vma || (msg->va != vma->vm_start) || vma->vm_private_data)
292 vma->vm_private_data = PRIV_VMA_LOCKED;
295 state.va = vma->vm_start;
297 state.domain = mmapcmd.dom;
299 rc = traverse_pages(mmapcmd.num, sizeof(struct privcmd_mmap_entry),
301 mmap_gfn_range, &state);
305 up_write(&mm->mmap_sem);
308 free_page_list(&pagelist);
313 struct mmap_batch_state {
316 struct vm_area_struct *vma;
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.
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;
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).
336 static int mmap_batch_fn(void *data, int nr, void *state)
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;
345 if (xen_feature(XENFEAT_auto_translated_physmap))
346 cur_pages = &pages[st->index];
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);
353 /* Adjust the global_error? */
356 st->global_error = -ENOENT;
358 /* Record that at least one error has happened. */
359 if (st->global_error == 0)
360 st->global_error = 1;
363 st->va += XEN_PAGE_SIZE * nr;
364 st->index += nr / XEN_PFN_PER_PAGE;
369 static int mmap_return_error(int err, struct mmap_batch_state *st)
373 if (st->version == 1) {
377 ret = get_user(gfn, st->user_gfn);
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).
385 gfn |= (err == -ENOENT) ?
386 PRIVCMD_MMAPBATCH_PAGED_ERROR :
387 PRIVCMD_MMAPBATCH_MFN_ERROR;
388 return __put_user(gfn, st->user_gfn++);
391 } else { /* st->version == 2 */
393 return __put_user(err, st->user_err++);
401 static int mmap_return_errors(void *data, int nr, void *state)
403 struct mmap_batch_state *st = state;
408 for (i = 0; i < nr; i++) {
409 ret = mmap_return_error(errs[i], st);
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
420 static int alloc_empty_pages(struct vm_area_struct *vma, int numpgs)
425 pages = kcalloc(numpgs, sizeof(pages[0]), GFP_KERNEL);
429 rc = alloc_xenballooned_pages(numpgs, pages);
431 pr_warn("%s Could not alloc %d pfns rc:%d\n", __func__,
436 BUG_ON(vma->vm_private_data != NULL);
437 vma->vm_private_data = pages;
442 static const struct vm_operations_struct privcmd_vm_ops;
444 static long privcmd_ioctl_mmap_batch(
445 struct file *file, void __user *udata, int version)
447 struct privcmd_data *data = file->private_data;
449 struct privcmd_mmapbatch_v2 m;
450 struct mm_struct *mm = current->mm;
451 struct vm_area_struct *vma;
452 unsigned long nr_pages;
454 struct mmap_batch_state state;
458 if (copy_from_user(&m, udata, sizeof(struct privcmd_mmapbatch)))
460 /* Returns per-frame error in m.arr. */
462 if (!access_ok(VERIFY_WRITE, m.arr, m.num * sizeof(*m.arr)))
466 if (copy_from_user(&m, udata, sizeof(struct privcmd_mmapbatch_v2)))
468 /* Returns per-frame error code in m.err. */
469 if (!access_ok(VERIFY_WRITE, m.err, m.num * (sizeof(*m.err))))
476 /* If restriction is in place, check the domid matches */
477 if (data->domid != DOMID_INVALID && data->domid != m.dom)
480 nr_pages = DIV_ROUND_UP(m.num, XEN_PFN_PER_PAGE);
481 if ((m.num <= 0) || (nr_pages > (LONG_MAX >> PAGE_SHIFT)))
484 ret = gather_array(&pagelist, m.num, sizeof(xen_pfn_t), m.arr);
488 if (list_empty(&pagelist)) {
494 /* Zero error array now to only copy back actual errors. */
495 if (clear_user(m.err, sizeof(int) * m.num)) {
501 down_write(&mm->mmap_sem);
503 vma = find_vma(mm, m.addr);
505 vma->vm_ops != &privcmd_vm_ops) {
511 * Caller must either:
513 * Map the whole VMA range, which will also allocate all the
514 * pages required for the auto_translated_physmap case.
518 * Map unmapped holes left from a previous map attempt (e.g.,
519 * because those foreign frames were previously paged out).
521 if (vma->vm_private_data == NULL) {
522 if (m.addr != vma->vm_start ||
523 m.addr + (nr_pages << PAGE_SHIFT) != vma->vm_end) {
527 if (xen_feature(XENFEAT_auto_translated_physmap)) {
528 ret = alloc_empty_pages(vma, nr_pages);
532 vma->vm_private_data = PRIV_VMA_LOCKED;
534 if (m.addr < vma->vm_start ||
535 m.addr + (nr_pages << PAGE_SHIFT) > vma->vm_end) {
539 if (privcmd_vma_range_is_mapped(vma, m.addr, nr_pages)) {
545 state.domain = m.dom;
549 state.global_error = 0;
550 state.version = version;
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));
557 up_write(&mm->mmap_sem);
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);
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))
574 free_page_list(&pagelist);
578 up_write(&mm->mmap_sem);
582 static int lock_pages(
583 struct privcmd_dm_op_buf kbufs[], unsigned int num,
584 struct page *pages[], unsigned int nr_pages)
588 for (i = 0; i < num; i++) {
589 unsigned int requested;
592 requested = DIV_ROUND_UP(
593 offset_in_page(kbufs[i].uptr) + kbufs[i].size,
595 if (requested > nr_pages)
598 pinned = get_user_pages_fast(
599 (unsigned long) kbufs[i].uptr,
600 requested, FOLL_WRITE, pages);
611 static void unlock_pages(struct page *pages[], unsigned int nr_pages)
618 for (i = 0; i < nr_pages; i++) {
624 static long privcmd_ioctl_dm_op(struct file *file, void __user *udata)
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;
635 if (copy_from_user(&kdata, udata, sizeof(kdata)))
638 /* If restriction is in place, check the domid matches */
639 if (data->domid != DOMID_INVALID && data->domid != kdata.dom)
645 if (kdata.num > privcmd_dm_op_max_num)
648 kbufs = kcalloc(kdata.num, sizeof(*kbufs), GFP_KERNEL);
652 if (copy_from_user(kbufs, kdata.ubufs,
653 sizeof(*kbufs) * kdata.num)) {
658 for (i = 0; i < kdata.num; i++) {
659 if (kbufs[i].size > privcmd_dm_op_buf_max_size) {
664 if (!access_ok(VERIFY_WRITE, kbufs[i].uptr,
670 nr_pages += DIV_ROUND_UP(
671 offset_in_page(kbufs[i].uptr) + kbufs[i].size,
675 pages = kcalloc(nr_pages, sizeof(*pages), GFP_KERNEL);
681 xbufs = kcalloc(kdata.num, sizeof(*xbufs), GFP_KERNEL);
687 rc = lock_pages(kbufs, kdata.num, pages, nr_pages);
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;
696 xen_preemptible_hcall_begin();
697 rc = HYPERVISOR_dm_op(kdata.dom, kdata.num, xbufs);
698 xen_preemptible_hcall_end();
701 unlock_pages(pages, nr_pages);
709 static long privcmd_ioctl_restrict(struct file *file, void __user *udata)
711 struct privcmd_data *data = file->private_data;
714 if (copy_from_user(&dom, udata, sizeof(dom)))
717 /* Set restriction to the specified domain, or check it matches */
718 if (data->domid == DOMID_INVALID)
720 else if (data->domid != dom)
727 struct mm_struct *mm;
733 static int remap_pfn_fn(pte_t *ptep, pgtable_t token, unsigned long addr,
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));
740 set_pte_at(r->mm, addr, ptep, pte);
746 static long privcmd_ioctl_mmap_resource(struct file *file, void __user *udata)
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;
756 if (copy_from_user(&kdata, udata, sizeof(kdata)))
759 /* If restriction is in place, check the domid matches */
760 if (data->domid != DOMID_INVALID && data->domid != kdata.dom)
763 down_write(&mm->mmap_sem);
765 vma = find_vma(mm, kdata.addr);
766 if (!vma || vma->vm_ops != &privcmd_vm_ops) {
771 pfns = kcalloc(kdata.num, sizeof(*pfns), GFP_KERNEL);
777 if (xen_feature(XENFEAT_auto_translated_physmap)) {
778 unsigned int nr = DIV_ROUND_UP(kdata.num, XEN_PFN_PER_PAGE);
782 rc = alloc_empty_pages(vma, nr);
786 pages = vma->vm_private_data;
787 for (i = 0; i < kdata.num; i++) {
789 page_to_xen_pfn(pages[i / XEN_PFN_PER_PAGE]);
791 pfns[i] = pfn + (i % XEN_PFN_PER_PAGE);
794 vma->vm_private_data = PRIV_VMA_LOCKED;
796 memset(&xdata, 0, sizeof(xdata));
797 xdata.domid = kdata.dom;
798 xdata.type = kdata.type;
800 xdata.frame = kdata.idx;
801 xdata.nr_frames = kdata.num;
802 set_xen_guest_handle(xdata.frame_list, pfns);
804 xen_preemptible_hcall_begin();
805 rc = HYPERVISOR_memory_op(XENMEM_acquire_resource, &xdata);
806 xen_preemptible_hcall_end();
811 if (xen_feature(XENFEAT_auto_translated_physmap)) {
812 struct remap_pfn r = {
814 .pages = vma->vm_private_data,
815 .prot = vma->vm_page_prot,
818 rc = apply_to_page_range(r.mm, kdata.addr,
819 kdata.num << PAGE_SHIFT,
823 (xdata.flags & XENMEM_rsrc_acq_caller_owned) ?
824 DOMID_SELF : kdata.dom;
827 num = xen_remap_domain_mfn_array(vma,
828 kdata.addr & PAGE_MASK,
829 pfns, kdata.num, (int *)pfns,
832 vma->vm_private_data);
835 else if (num != kdata.num) {
838 for (i = 0; i < num; i++) {
848 up_write(&mm->mmap_sem);
854 static long privcmd_ioctl(struct file *file,
855 unsigned int cmd, unsigned long data)
858 void __user *udata = (void __user *) data;
861 case IOCTL_PRIVCMD_HYPERCALL:
862 ret = privcmd_ioctl_hypercall(file, udata);
865 case IOCTL_PRIVCMD_MMAP:
866 ret = privcmd_ioctl_mmap(file, udata);
869 case IOCTL_PRIVCMD_MMAPBATCH:
870 ret = privcmd_ioctl_mmap_batch(file, udata, 1);
873 case IOCTL_PRIVCMD_MMAPBATCH_V2:
874 ret = privcmd_ioctl_mmap_batch(file, udata, 2);
877 case IOCTL_PRIVCMD_DM_OP:
878 ret = privcmd_ioctl_dm_op(file, udata);
881 case IOCTL_PRIVCMD_RESTRICT:
882 ret = privcmd_ioctl_restrict(file, udata);
885 case IOCTL_PRIVCMD_MMAP_RESOURCE:
886 ret = privcmd_ioctl_mmap_resource(file, udata);
896 static int privcmd_open(struct inode *ino, struct file *file)
898 struct privcmd_data *data = kzalloc(sizeof(*data), GFP_KERNEL);
903 /* DOMID_INVALID implies no restriction */
904 data->domid = DOMID_INVALID;
906 file->private_data = data;
910 static int privcmd_release(struct inode *ino, struct file *file)
912 struct privcmd_data *data = file->private_data;
918 static void privcmd_close(struct vm_area_struct *vma)
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;
925 if (!xen_feature(XENFEAT_auto_translated_physmap) || !numpgs || !pages)
928 rc = xen_unmap_domain_gfn_range(vma, numgfns, pages);
930 free_xenballooned_pages(numpgs, pages);
932 pr_crit("unable to unmap MFN range: leaking %d pages. rc=%d\n",
937 static vm_fault_t privcmd_fault(struct vm_fault *vmf)
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);
943 return VM_FAULT_SIGBUS;
946 static const struct vm_operations_struct privcmd_vm_ops = {
947 .close = privcmd_close,
948 .fault = privcmd_fault
951 static int privcmd_mmap(struct file *file, struct vm_area_struct *vma)
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;
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.
968 static int is_mapped_fn(pte_t *pte, struct page *pmd_page,
969 unsigned long addr, void *data)
971 return pte_none(*pte) ? 0 : -EBUSY;
974 static int privcmd_vma_range_is_mapped(
975 struct vm_area_struct *vma,
977 unsigned long nr_pages)
979 return apply_to_page_range(vma->vm_mm, addr, nr_pages << PAGE_SHIFT,
980 is_mapped_fn, NULL) != 0;
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,
990 EXPORT_SYMBOL_GPL(xen_privcmd_fops);
992 static struct miscdevice privcmd_dev = {
993 .minor = MISC_DYNAMIC_MINOR,
994 .name = "xen/privcmd",
995 .fops = &xen_privcmd_fops,
998 static int __init privcmd_init(void)
1005 err = misc_register(&privcmd_dev);
1007 pr_err("Could not register Xen privcmd device\n");
1011 err = misc_register(&xen_privcmdbuf_dev);
1013 pr_err("Could not register Xen hypercall-buf device\n");
1014 misc_deregister(&privcmd_dev);
1021 static void __exit privcmd_exit(void)
1023 misc_deregister(&privcmd_dev);
1024 misc_deregister(&xen_privcmdbuf_dev);
1027 module_init(privcmd_init);
1028 module_exit(privcmd_exit);