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>
26 #include <asm/pgalloc.h>
27 #include <asm/pgtable.h>
29 #include <asm/xen/hypervisor.h>
30 #include <asm/xen/hypercall.h>
33 #include <xen/privcmd.h>
34 #include <xen/interface/xen.h>
35 #include <xen/features.h>
37 #include <xen/xen-ops.h>
38 #include <xen/balloon.h>
42 MODULE_LICENSE("GPL");
44 #define PRIV_VMA_LOCKED ((void *)1)
46 static int privcmd_vma_range_is_mapped(
47 struct vm_area_struct *vma,
49 unsigned long nr_pages);
51 static long privcmd_ioctl_hypercall(void __user *udata)
53 struct privcmd_hypercall hypercall;
56 if (copy_from_user(&hypercall, udata, sizeof(hypercall)))
59 xen_preemptible_hcall_begin();
60 ret = privcmd_call(hypercall.op,
61 hypercall.arg[0], hypercall.arg[1],
62 hypercall.arg[2], hypercall.arg[3],
64 xen_preemptible_hcall_end();
69 static void free_page_list(struct list_head *pages)
73 list_for_each_entry_safe(p, n, pages, lru)
76 INIT_LIST_HEAD(pages);
80 * Given an array of items in userspace, return a list of pages
81 * containing the data. If copying fails, either because of memory
82 * allocation failure or a problem reading user memory, return an
83 * error code; its up to the caller to dispose of any partial list.
85 static int gather_array(struct list_head *pagelist,
86 unsigned nelem, size_t size,
87 const void __user *data)
97 pagedata = NULL; /* quiet, gcc */
99 if (pageidx > PAGE_SIZE-size) {
100 struct page *page = alloc_page(GFP_KERNEL);
106 pagedata = page_address(page);
108 list_add_tail(&page->lru, pagelist);
113 if (copy_from_user(pagedata + pageidx, data, size))
127 * Call function "fn" on each element of the array fragmented
128 * over a list of pages.
130 static int traverse_pages(unsigned nelem, size_t size,
131 struct list_head *pos,
132 int (*fn)(void *data, void *state),
139 BUG_ON(size > PAGE_SIZE);
142 pagedata = NULL; /* hush, gcc */
145 if (pageidx > PAGE_SIZE-size) {
148 page = list_entry(pos, struct page, lru);
149 pagedata = page_address(page);
153 ret = (*fn)(pagedata + pageidx, state);
163 * Similar to traverse_pages, but use each page as a "block" of
164 * data to be processed as one unit.
166 static int traverse_pages_block(unsigned nelem, size_t size,
167 struct list_head *pos,
168 int (*fn)(void *data, int nr, void *state),
175 BUG_ON(size > PAGE_SIZE);
180 int nr = (PAGE_SIZE/size);
185 page = list_entry(pos, struct page, lru);
186 pagedata = page_address(page);
187 ret = (*fn)(pagedata, nr, state);
196 struct mmap_gfn_state {
198 struct vm_area_struct *vma;
202 static int mmap_gfn_range(void *data, void *state)
204 struct privcmd_mmap_entry *msg = data;
205 struct mmap_gfn_state *st = state;
206 struct vm_area_struct *vma = st->vma;
209 /* Do not allow range to wrap the address space. */
210 if ((msg->npages > (LONG_MAX >> PAGE_SHIFT)) ||
211 ((unsigned long)(msg->npages << PAGE_SHIFT) >= -st->va))
214 /* Range chunks must be contiguous in va space. */
215 if ((msg->va != st->va) ||
216 ((msg->va+(msg->npages<<PAGE_SHIFT)) > vma->vm_end))
219 rc = xen_remap_domain_gfn_range(vma,
221 msg->mfn, msg->npages,
227 st->va += msg->npages << PAGE_SHIFT;
232 static long privcmd_ioctl_mmap(void __user *udata)
234 struct privcmd_mmap mmapcmd;
235 struct mm_struct *mm = current->mm;
236 struct vm_area_struct *vma;
239 struct mmap_gfn_state state;
241 /* We only support privcmd_ioctl_mmap_batch for auto translated. */
242 if (xen_feature(XENFEAT_auto_translated_physmap))
245 if (copy_from_user(&mmapcmd, udata, sizeof(mmapcmd)))
248 rc = gather_array(&pagelist,
249 mmapcmd.num, sizeof(struct privcmd_mmap_entry),
252 if (rc || list_empty(&pagelist))
255 down_write(&mm->mmap_sem);
258 struct page *page = list_first_entry(&pagelist,
260 struct privcmd_mmap_entry *msg = page_address(page);
262 vma = find_vma(mm, msg->va);
265 if (!vma || (msg->va != vma->vm_start) || vma->vm_private_data)
267 vma->vm_private_data = PRIV_VMA_LOCKED;
270 state.va = vma->vm_start;
272 state.domain = mmapcmd.dom;
274 rc = traverse_pages(mmapcmd.num, sizeof(struct privcmd_mmap_entry),
276 mmap_gfn_range, &state);
280 up_write(&mm->mmap_sem);
283 free_page_list(&pagelist);
288 struct mmap_batch_state {
291 struct vm_area_struct *vma;
295 * 1 if at least one error has happened (and no
296 * -ENOENT errors have happened)
297 * -ENOENT if at least 1 -ENOENT has happened.
302 /* User-space gfn array to store errors in the second pass for V1. */
303 xen_pfn_t __user *user_gfn;
304 /* User-space int array to store errors in the second pass for V2. */
305 int __user *user_err;
308 /* auto translated dom0 note: if domU being created is PV, then gfn is
309 * mfn(addr on bus). If it's auto xlated, then gfn is pfn (input to HAP).
311 static int mmap_batch_fn(void *data, int nr, void *state)
313 xen_pfn_t *gfnp = data;
314 struct mmap_batch_state *st = state;
315 struct vm_area_struct *vma = st->vma;
316 struct page **pages = vma->vm_private_data;
317 struct page **cur_pages = NULL;
320 if (xen_feature(XENFEAT_auto_translated_physmap))
321 cur_pages = &pages[st->index];
324 ret = xen_remap_domain_gfn_array(st->vma, st->va & PAGE_MASK, gfnp, nr,
325 (int *)gfnp, st->vma->vm_page_prot,
326 st->domain, cur_pages);
328 /* Adjust the global_error? */
331 st->global_error = -ENOENT;
333 /* Record that at least one error has happened. */
334 if (st->global_error == 0)
335 st->global_error = 1;
338 st->va += PAGE_SIZE * nr;
344 static int mmap_return_error(int err, struct mmap_batch_state *st)
348 if (st->version == 1) {
352 ret = get_user(gfn, st->user_gfn);
356 * V1 encodes the error codes in the 32bit top
357 * nibble of the gfn (with its known
358 * limitations vis-a-vis 64 bit callers).
360 gfn |= (err == -ENOENT) ?
361 PRIVCMD_MMAPBATCH_PAGED_ERROR :
362 PRIVCMD_MMAPBATCH_MFN_ERROR;
363 return __put_user(gfn, st->user_gfn++);
366 } else { /* st->version == 2 */
368 return __put_user(err, st->user_err++);
376 static int mmap_return_errors(void *data, int nr, void *state)
378 struct mmap_batch_state *st = state;
383 for (i = 0; i < nr; i++) {
384 ret = mmap_return_error(errs[i], st);
391 /* Allocate pfns that are then mapped with gfns from foreign domid. Update
392 * the vma with the page info to use later.
393 * Returns: 0 if success, otherwise -errno
395 static int alloc_empty_pages(struct vm_area_struct *vma, int numpgs)
400 pages = kcalloc(numpgs, sizeof(pages[0]), GFP_KERNEL);
404 rc = alloc_xenballooned_pages(numpgs, pages);
406 pr_warn("%s Could not alloc %d pfns rc:%d\n", __func__,
411 BUG_ON(vma->vm_private_data != NULL);
412 vma->vm_private_data = pages;
417 static const struct vm_operations_struct privcmd_vm_ops;
419 static long privcmd_ioctl_mmap_batch(void __user *udata, int version)
422 struct privcmd_mmapbatch_v2 m;
423 struct mm_struct *mm = current->mm;
424 struct vm_area_struct *vma;
425 unsigned long nr_pages;
427 struct mmap_batch_state state;
431 if (copy_from_user(&m, udata, sizeof(struct privcmd_mmapbatch)))
433 /* Returns per-frame error in m.arr. */
435 if (!access_ok(VERIFY_WRITE, m.arr, m.num * sizeof(*m.arr)))
439 if (copy_from_user(&m, udata, sizeof(struct privcmd_mmapbatch_v2)))
441 /* Returns per-frame error code in m.err. */
442 if (!access_ok(VERIFY_WRITE, m.err, m.num * (sizeof(*m.err))))
449 nr_pages = DIV_ROUND_UP(m.num, XEN_PFN_PER_PAGE);
450 if ((m.num <= 0) || (nr_pages > (LONG_MAX >> PAGE_SHIFT)))
453 ret = gather_array(&pagelist, m.num, sizeof(xen_pfn_t), m.arr);
457 if (list_empty(&pagelist)) {
463 /* Zero error array now to only copy back actual errors. */
464 if (clear_user(m.err, sizeof(int) * m.num)) {
470 down_write(&mm->mmap_sem);
472 vma = find_vma(mm, m.addr);
474 vma->vm_ops != &privcmd_vm_ops) {
480 * Caller must either:
482 * Map the whole VMA range, which will also allocate all the
483 * pages required for the auto_translated_physmap case.
487 * Map unmapped holes left from a previous map attempt (e.g.,
488 * because those foreign frames were previously paged out).
490 if (vma->vm_private_data == NULL) {
491 if (m.addr != vma->vm_start ||
492 m.addr + (nr_pages << PAGE_SHIFT) != vma->vm_end) {
496 if (xen_feature(XENFEAT_auto_translated_physmap)) {
497 ret = alloc_empty_pages(vma, nr_pages);
501 vma->vm_private_data = PRIV_VMA_LOCKED;
503 if (m.addr < vma->vm_start ||
504 m.addr + (nr_pages << PAGE_SHIFT) > vma->vm_end) {
508 if (privcmd_vma_range_is_mapped(vma, m.addr, nr_pages)) {
514 state.domain = m.dom;
518 state.global_error = 0;
519 state.version = version;
521 BUILD_BUG_ON(((PAGE_SIZE / sizeof(xen_pfn_t)) % XEN_PFN_PER_PAGE) != 0);
522 /* mmap_batch_fn guarantees ret == 0 */
523 BUG_ON(traverse_pages_block(m.num, sizeof(xen_pfn_t),
524 &pagelist, mmap_batch_fn, &state));
526 up_write(&mm->mmap_sem);
528 if (state.global_error) {
529 /* Write back errors in second pass. */
530 state.user_gfn = (xen_pfn_t *)m.arr;
531 state.user_err = m.err;
532 ret = traverse_pages_block(m.num, sizeof(xen_pfn_t),
533 &pagelist, mmap_return_errors, &state);
537 /* If we have not had any EFAULT-like global errors then set the global
538 * error to -ENOENT if necessary. */
539 if ((ret == 0) && (state.global_error == -ENOENT))
543 free_page_list(&pagelist);
547 up_write(&mm->mmap_sem);
551 static long privcmd_ioctl(struct file *file,
552 unsigned int cmd, unsigned long data)
555 void __user *udata = (void __user *) data;
558 case IOCTL_PRIVCMD_HYPERCALL:
559 ret = privcmd_ioctl_hypercall(udata);
562 case IOCTL_PRIVCMD_MMAP:
563 ret = privcmd_ioctl_mmap(udata);
566 case IOCTL_PRIVCMD_MMAPBATCH:
567 ret = privcmd_ioctl_mmap_batch(udata, 1);
570 case IOCTL_PRIVCMD_MMAPBATCH_V2:
571 ret = privcmd_ioctl_mmap_batch(udata, 2);
582 static void privcmd_close(struct vm_area_struct *vma)
584 struct page **pages = vma->vm_private_data;
585 int numpgs = (vma->vm_end - vma->vm_start) >> PAGE_SHIFT;
586 int numgfns = (vma->vm_end - vma->vm_start) >> XEN_PAGE_SHIFT;
589 if (!xen_feature(XENFEAT_auto_translated_physmap) || !numpgs || !pages)
592 rc = xen_unmap_domain_gfn_range(vma, numgfns, pages);
594 free_xenballooned_pages(numpgs, pages);
596 pr_crit("unable to unmap MFN range: leaking %d pages. rc=%d\n",
601 static int privcmd_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
603 printk(KERN_DEBUG "privcmd_fault: vma=%p %lx-%lx, pgoff=%lx, uv=%p\n",
604 vma, vma->vm_start, vma->vm_end,
605 vmf->pgoff, vmf->virtual_address);
607 return VM_FAULT_SIGBUS;
610 static const struct vm_operations_struct privcmd_vm_ops = {
611 .close = privcmd_close,
612 .fault = privcmd_fault
615 static int privcmd_mmap(struct file *file, struct vm_area_struct *vma)
617 /* DONTCOPY is essential for Xen because copy_page_range doesn't know
618 * how to recreate these mappings */
619 vma->vm_flags |= VM_IO | VM_PFNMAP | VM_DONTCOPY |
620 VM_DONTEXPAND | VM_DONTDUMP;
621 vma->vm_ops = &privcmd_vm_ops;
622 vma->vm_private_data = NULL;
628 * For MMAPBATCH*. This allows asserting the singleshot mapping
629 * on a per pfn/pte basis. Mapping calls that fail with ENOENT
630 * can be then retried until success.
632 static int is_mapped_fn(pte_t *pte, struct page *pmd_page,
633 unsigned long addr, void *data)
635 return pte_none(*pte) ? 0 : -EBUSY;
638 static int privcmd_vma_range_is_mapped(
639 struct vm_area_struct *vma,
641 unsigned long nr_pages)
643 return apply_to_page_range(vma->vm_mm, addr, nr_pages << PAGE_SHIFT,
644 is_mapped_fn, NULL) != 0;
647 const struct file_operations xen_privcmd_fops = {
648 .owner = THIS_MODULE,
649 .unlocked_ioctl = privcmd_ioctl,
650 .mmap = privcmd_mmap,
652 EXPORT_SYMBOL_GPL(xen_privcmd_fops);
654 static struct miscdevice privcmd_dev = {
655 .minor = MISC_DYNAMIC_MINOR,
656 .name = "xen/privcmd",
657 .fops = &xen_privcmd_fops,
660 static int __init privcmd_init(void)
667 err = misc_register(&privcmd_dev);
669 pr_err("Could not register Xen privcmd device\n");
675 static void __exit privcmd_exit(void)
677 misc_deregister(&privcmd_dev);
680 module_init(privcmd_init);
681 module_exit(privcmd_exit);