1 // SPDX-License-Identifier: GPL-2.0-only
2 /******************************************************************************
5 * Interface to privileged domain-0 commands.
7 * Copyright (c) 2002-2004, K A Fraser, B Dragovic
10 #define pr_fmt(fmt) "xen:" KBUILD_MODNAME ": " fmt
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>
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>
28 #include <asm/pgalloc.h>
29 #include <asm/pgtable.h>
31 #include <asm/xen/hypervisor.h>
32 #include <asm/xen/hypercall.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>
41 #include <xen/xen-ops.h>
42 #include <xen/balloon.h>
46 MODULE_LICENSE("GPL");
48 #define PRIV_VMA_LOCKED ((void *)1)
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");
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,
58 MODULE_PARM_DESC(dm_op_buf_max_size,
59 "Maximum size of a dm_op hypercall buffer");
65 static int privcmd_vma_range_is_mapped(
66 struct vm_area_struct *vma,
68 unsigned long nr_pages);
70 static long privcmd_ioctl_hypercall(struct file *file, void __user *udata)
72 struct privcmd_data *data = file->private_data;
73 struct privcmd_hypercall hypercall;
76 /* Disallow arbitrary hypercalls if restricted */
77 if (data->domid != DOMID_INVALID)
80 if (copy_from_user(&hypercall, udata, sizeof(hypercall)))
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],
88 xen_preemptible_hcall_end();
93 static void free_page_list(struct list_head *pages)
97 list_for_each_entry_safe(p, n, pages, lru)
100 INIT_LIST_HEAD(pages);
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.
109 static int gather_array(struct list_head *pagelist,
110 unsigned nelem, size_t size,
111 const void __user *data)
117 if (size > PAGE_SIZE)
121 pagedata = NULL; /* quiet, gcc */
123 if (pageidx > PAGE_SIZE-size) {
124 struct page *page = alloc_page(GFP_KERNEL);
130 pagedata = page_address(page);
132 list_add_tail(&page->lru, pagelist);
137 if (copy_from_user(pagedata + pageidx, data, size))
151 * Call function "fn" on each element of the array fragmented
152 * over a list of pages.
154 static int traverse_pages(unsigned nelem, size_t size,
155 struct list_head *pos,
156 int (*fn)(void *data, void *state),
163 BUG_ON(size > PAGE_SIZE);
166 pagedata = NULL; /* hush, gcc */
169 if (pageidx > PAGE_SIZE-size) {
172 page = list_entry(pos, struct page, lru);
173 pagedata = page_address(page);
177 ret = (*fn)(pagedata + pageidx, state);
187 * Similar to traverse_pages, but use each page as a "block" of
188 * data to be processed as one unit.
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),
198 BUG_ON(size > PAGE_SIZE);
201 int nr = (PAGE_SIZE/size);
206 page = list_entry(pos, struct page, lru);
207 pagedata = page_address(page);
208 ret = (*fn)(pagedata, nr, state);
217 struct mmap_gfn_state {
219 struct vm_area_struct *vma;
223 static int mmap_gfn_range(void *data, void *state)
225 struct privcmd_mmap_entry *msg = data;
226 struct mmap_gfn_state *st = state;
227 struct vm_area_struct *vma = st->vma;
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))
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))
240 rc = xen_remap_domain_gfn_range(vma,
242 msg->mfn, msg->npages,
248 st->va += msg->npages << PAGE_SHIFT;
253 static long privcmd_ioctl_mmap(struct file *file, void __user *udata)
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;
261 struct mmap_gfn_state state;
263 /* We only support privcmd_ioctl_mmap_batch for auto translated. */
264 if (xen_feature(XENFEAT_auto_translated_physmap))
267 if (copy_from_user(&mmapcmd, udata, sizeof(mmapcmd)))
270 /* If restriction is in place, check the domid matches */
271 if (data->domid != DOMID_INVALID && data->domid != mmapcmd.dom)
274 rc = gather_array(&pagelist,
275 mmapcmd.num, sizeof(struct privcmd_mmap_entry),
278 if (rc || list_empty(&pagelist))
281 down_write(&mm->mmap_sem);
284 struct page *page = list_first_entry(&pagelist,
286 struct privcmd_mmap_entry *msg = page_address(page);
288 vma = find_vma(mm, msg->va);
291 if (!vma || (msg->va != vma->vm_start) || vma->vm_private_data)
293 vma->vm_private_data = PRIV_VMA_LOCKED;
296 state.va = vma->vm_start;
298 state.domain = mmapcmd.dom;
300 rc = traverse_pages(mmapcmd.num, sizeof(struct privcmd_mmap_entry),
302 mmap_gfn_range, &state);
306 up_write(&mm->mmap_sem);
309 free_page_list(&pagelist);
314 struct mmap_batch_state {
317 struct vm_area_struct *vma;
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.
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;
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).
337 static int mmap_batch_fn(void *data, int nr, void *state)
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;
346 if (xen_feature(XENFEAT_auto_translated_physmap))
347 cur_pages = &pages[st->index];
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);
354 /* Adjust the global_error? */
357 st->global_error = -ENOENT;
359 /* Record that at least one error has happened. */
360 if (st->global_error == 0)
361 st->global_error = 1;
364 st->va += XEN_PAGE_SIZE * nr;
365 st->index += nr / XEN_PFN_PER_PAGE;
370 static int mmap_return_error(int err, struct mmap_batch_state *st)
374 if (st->version == 1) {
378 ret = get_user(gfn, st->user_gfn);
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).
386 gfn |= (err == -ENOENT) ?
387 PRIVCMD_MMAPBATCH_PAGED_ERROR :
388 PRIVCMD_MMAPBATCH_MFN_ERROR;
389 return __put_user(gfn, st->user_gfn++);
392 } else { /* st->version == 2 */
394 return __put_user(err, st->user_err++);
402 static int mmap_return_errors(void *data, int nr, void *state)
404 struct mmap_batch_state *st = state;
409 for (i = 0; i < nr; i++) {
410 ret = mmap_return_error(errs[i], st);
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
421 static int alloc_empty_pages(struct vm_area_struct *vma, int numpgs)
426 pages = kcalloc(numpgs, sizeof(pages[0]), GFP_KERNEL);
430 rc = alloc_xenballooned_pages(numpgs, pages);
432 pr_warn("%s Could not alloc %d pfns rc:%d\n", __func__,
437 BUG_ON(vma->vm_private_data != NULL);
438 vma->vm_private_data = pages;
443 static const struct vm_operations_struct privcmd_vm_ops;
445 static long privcmd_ioctl_mmap_batch(
446 struct file *file, void __user *udata, int version)
448 struct privcmd_data *data = file->private_data;
450 struct privcmd_mmapbatch_v2 m;
451 struct mm_struct *mm = current->mm;
452 struct vm_area_struct *vma;
453 unsigned long nr_pages;
455 struct mmap_batch_state state;
459 if (copy_from_user(&m, udata, sizeof(struct privcmd_mmapbatch)))
461 /* Returns per-frame error in m.arr. */
463 if (!access_ok(m.arr, m.num * sizeof(*m.arr)))
467 if (copy_from_user(&m, udata, sizeof(struct privcmd_mmapbatch_v2)))
469 /* Returns per-frame error code in m.err. */
470 if (!access_ok(m.err, m.num * (sizeof(*m.err))))
477 /* If restriction is in place, check the domid matches */
478 if (data->domid != DOMID_INVALID && data->domid != m.dom)
481 nr_pages = DIV_ROUND_UP(m.num, XEN_PFN_PER_PAGE);
482 if ((m.num <= 0) || (nr_pages > (LONG_MAX >> PAGE_SHIFT)))
485 ret = gather_array(&pagelist, m.num, sizeof(xen_pfn_t), m.arr);
489 if (list_empty(&pagelist)) {
495 /* Zero error array now to only copy back actual errors. */
496 if (clear_user(m.err, sizeof(int) * m.num)) {
502 down_write(&mm->mmap_sem);
504 vma = find_vma(mm, m.addr);
506 vma->vm_ops != &privcmd_vm_ops) {
512 * Caller must either:
514 * Map the whole VMA range, which will also allocate all the
515 * pages required for the auto_translated_physmap case.
519 * Map unmapped holes left from a previous map attempt (e.g.,
520 * because those foreign frames were previously paged out).
522 if (vma->vm_private_data == NULL) {
523 if (m.addr != vma->vm_start ||
524 m.addr + (nr_pages << PAGE_SHIFT) != vma->vm_end) {
528 if (xen_feature(XENFEAT_auto_translated_physmap)) {
529 ret = alloc_empty_pages(vma, nr_pages);
533 vma->vm_private_data = PRIV_VMA_LOCKED;
535 if (m.addr < vma->vm_start ||
536 m.addr + (nr_pages << PAGE_SHIFT) > vma->vm_end) {
540 if (privcmd_vma_range_is_mapped(vma, m.addr, nr_pages)) {
546 state.domain = m.dom;
550 state.global_error = 0;
551 state.version = version;
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));
558 up_write(&mm->mmap_sem);
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);
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))
575 free_page_list(&pagelist);
579 up_write(&mm->mmap_sem);
583 static int lock_pages(
584 struct privcmd_dm_op_buf kbufs[], unsigned int num,
585 struct page *pages[], unsigned int nr_pages)
589 for (i = 0; i < num; i++) {
590 unsigned int requested;
593 requested = DIV_ROUND_UP(
594 offset_in_page(kbufs[i].uptr) + kbufs[i].size,
596 if (requested > nr_pages)
599 pinned = get_user_pages_fast(
600 (unsigned long) kbufs[i].uptr,
601 requested, FOLL_WRITE, pages);
612 static void unlock_pages(struct page *pages[], unsigned int nr_pages)
619 for (i = 0; i < nr_pages; i++) {
625 static long privcmd_ioctl_dm_op(struct file *file, void __user *udata)
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;
636 if (copy_from_user(&kdata, udata, sizeof(kdata)))
639 /* If restriction is in place, check the domid matches */
640 if (data->domid != DOMID_INVALID && data->domid != kdata.dom)
646 if (kdata.num > privcmd_dm_op_max_num)
649 kbufs = kcalloc(kdata.num, sizeof(*kbufs), GFP_KERNEL);
653 if (copy_from_user(kbufs, kdata.ubufs,
654 sizeof(*kbufs) * kdata.num)) {
659 for (i = 0; i < kdata.num; i++) {
660 if (kbufs[i].size > privcmd_dm_op_buf_max_size) {
665 if (!access_ok(kbufs[i].uptr,
671 nr_pages += DIV_ROUND_UP(
672 offset_in_page(kbufs[i].uptr) + kbufs[i].size,
676 pages = kcalloc(nr_pages, sizeof(*pages), GFP_KERNEL);
682 xbufs = kcalloc(kdata.num, sizeof(*xbufs), GFP_KERNEL);
688 rc = lock_pages(kbufs, kdata.num, pages, nr_pages);
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;
697 xen_preemptible_hcall_begin();
698 rc = HYPERVISOR_dm_op(kdata.dom, kdata.num, xbufs);
699 xen_preemptible_hcall_end();
702 unlock_pages(pages, nr_pages);
710 static long privcmd_ioctl_restrict(struct file *file, void __user *udata)
712 struct privcmd_data *data = file->private_data;
715 if (copy_from_user(&dom, udata, sizeof(dom)))
718 /* Set restriction to the specified domain, or check it matches */
719 if (data->domid == DOMID_INVALID)
721 else if (data->domid != dom)
728 struct mm_struct *mm;
734 static int remap_pfn_fn(pte_t *ptep, unsigned long addr, void *data)
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, unsigned long addr, void *data)
970 return pte_none(*pte) ? 0 : -EBUSY;
973 static int privcmd_vma_range_is_mapped(
974 struct vm_area_struct *vma,
976 unsigned long nr_pages)
978 return apply_to_page_range(vma->vm_mm, addr, nr_pages << PAGE_SHIFT,
979 is_mapped_fn, NULL) != 0;
982 const struct file_operations xen_privcmd_fops = {
983 .owner = THIS_MODULE,
984 .unlocked_ioctl = privcmd_ioctl,
985 .open = privcmd_open,
986 .release = privcmd_release,
987 .mmap = privcmd_mmap,
989 EXPORT_SYMBOL_GPL(xen_privcmd_fops);
991 static struct miscdevice privcmd_dev = {
992 .minor = MISC_DYNAMIC_MINOR,
993 .name = "xen/privcmd",
994 .fops = &xen_privcmd_fops,
997 static int __init privcmd_init(void)
1004 err = misc_register(&privcmd_dev);
1006 pr_err("Could not register Xen privcmd device\n");
1010 err = misc_register(&xen_privcmdbuf_dev);
1012 pr_err("Could not register Xen hypercall-buf device\n");
1013 misc_deregister(&privcmd_dev);
1020 static void __exit privcmd_exit(void)
1022 misc_deregister(&privcmd_dev);
1023 misc_deregister(&xen_privcmdbuf_dev);
1026 module_init(privcmd_init);
1027 module_exit(privcmd_exit);