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 ret = privcmd_call(hypercall.op,
60 hypercall.arg[0], hypercall.arg[1],
61 hypercall.arg[2], hypercall.arg[3],
67 static void free_page_list(struct list_head *pages)
71 list_for_each_entry_safe(p, n, pages, lru)
74 INIT_LIST_HEAD(pages);
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.
83 static int gather_array(struct list_head *pagelist,
84 unsigned nelem, size_t size,
85 const void __user *data)
95 pagedata = NULL; /* quiet, gcc */
97 if (pageidx > PAGE_SIZE-size) {
98 struct page *page = alloc_page(GFP_KERNEL);
104 pagedata = page_address(page);
106 list_add_tail(&page->lru, pagelist);
111 if (copy_from_user(pagedata + pageidx, data, size))
125 * Call function "fn" on each element of the array fragmented
126 * over a list of pages.
128 static int traverse_pages(unsigned nelem, size_t size,
129 struct list_head *pos,
130 int (*fn)(void *data, void *state),
137 BUG_ON(size > PAGE_SIZE);
140 pagedata = NULL; /* hush, gcc */
143 if (pageidx > PAGE_SIZE-size) {
146 page = list_entry(pos, struct page, lru);
147 pagedata = page_address(page);
151 ret = (*fn)(pagedata + pageidx, state);
160 struct mmap_mfn_state {
162 struct vm_area_struct *vma;
166 static int mmap_mfn_range(void *data, void *state)
168 struct privcmd_mmap_entry *msg = data;
169 struct mmap_mfn_state *st = state;
170 struct vm_area_struct *vma = st->vma;
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))
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))
183 rc = xen_remap_domain_mfn_range(vma,
185 msg->mfn, msg->npages,
191 st->va += msg->npages << PAGE_SHIFT;
196 static long privcmd_ioctl_mmap(void __user *udata)
198 struct privcmd_mmap mmapcmd;
199 struct mm_struct *mm = current->mm;
200 struct vm_area_struct *vma;
203 struct mmap_mfn_state state;
205 /* We only support privcmd_ioctl_mmap_batch for auto translated. */
206 if (xen_feature(XENFEAT_auto_translated_physmap))
209 if (copy_from_user(&mmapcmd, udata, sizeof(mmapcmd)))
212 rc = gather_array(&pagelist,
213 mmapcmd.num, sizeof(struct privcmd_mmap_entry),
216 if (rc || list_empty(&pagelist))
219 down_write(&mm->mmap_sem);
222 struct page *page = list_first_entry(&pagelist,
224 struct privcmd_mmap_entry *msg = page_address(page);
226 vma = find_vma(mm, msg->va);
229 if (!vma || (msg->va != vma->vm_start) || vma->vm_private_data)
231 vma->vm_private_data = PRIV_VMA_LOCKED;
234 state.va = vma->vm_start;
236 state.domain = mmapcmd.dom;
238 rc = traverse_pages(mmapcmd.num, sizeof(struct privcmd_mmap_entry),
240 mmap_mfn_range, &state);
244 up_write(&mm->mmap_sem);
247 free_page_list(&pagelist);
252 struct mmap_batch_state {
255 struct vm_area_struct *vma;
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.
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;
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).
275 static int mmap_batch_fn(void *data, void *state)
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;
284 if (xen_feature(XENFEAT_auto_translated_physmap))
285 cur_page = pages[st->index++];
287 ret = xen_remap_domain_mfn_range(st->vma, st->va & PAGE_MASK, *mfnp, 1,
288 st->vma->vm_page_prot, st->domain,
291 /* Store error code for second pass. */
292 if (st->version == 1) {
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).
298 *mfnp |= (ret == -ENOENT) ?
299 PRIVCMD_MMAPBATCH_PAGED_ERROR :
300 PRIVCMD_MMAPBATCH_MFN_ERROR;
302 } else { /* st->version == 2 */
303 *((int *) mfnp) = ret;
306 /* And see if it affects the global_error. */
309 st->global_error = -ENOENT;
311 /* Record that at least one error has happened. */
312 if (st->global_error == 0)
313 st->global_error = 1;
321 static int mmap_return_errors(void *data, void *state)
323 struct mmap_batch_state *st = state;
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++);
331 } else { /* st->version == 2 */
332 int err = *((int *) data);
334 return __put_user(err, st->user_err++);
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
346 static int alloc_empty_pages(struct vm_area_struct *vma, int numpgs)
351 pages = kcalloc(numpgs, sizeof(pages[0]), GFP_KERNEL);
355 rc = alloc_xenballooned_pages(numpgs, pages, 0);
357 pr_warn("%s Could not alloc %d pfns rc:%d\n", __func__,
362 BUG_ON(vma->vm_private_data != NULL);
363 vma->vm_private_data = pages;
368 static struct vm_operations_struct privcmd_vm_ops;
370 static long privcmd_ioctl_mmap_batch(void __user *udata, int version)
373 struct privcmd_mmapbatch_v2 m;
374 struct mm_struct *mm = current->mm;
375 struct vm_area_struct *vma;
376 unsigned long nr_pages;
378 struct mmap_batch_state state;
382 if (copy_from_user(&m, udata, sizeof(struct privcmd_mmapbatch)))
384 /* Returns per-frame error in m.arr. */
386 if (!access_ok(VERIFY_WRITE, m.arr, m.num * sizeof(*m.arr)))
390 if (copy_from_user(&m, udata, sizeof(struct privcmd_mmapbatch_v2)))
392 /* Returns per-frame error code in m.err. */
393 if (!access_ok(VERIFY_WRITE, m.err, m.num * (sizeof(*m.err))))
401 if ((m.num <= 0) || (nr_pages > (LONG_MAX >> PAGE_SHIFT)))
404 ret = gather_array(&pagelist, m.num, sizeof(xen_pfn_t), m.arr);
408 if (list_empty(&pagelist)) {
414 /* Zero error array now to only copy back actual errors. */
415 if (clear_user(m.err, sizeof(int) * m.num)) {
421 down_write(&mm->mmap_sem);
423 vma = find_vma(mm, m.addr);
425 vma->vm_ops != &privcmd_vm_ops) {
431 * Caller must either:
433 * Map the whole VMA range, which will also allocate all the
434 * pages required for the auto_translated_physmap case.
438 * Map unmapped holes left from a previous map attempt (e.g.,
439 * because those foreign frames were previously paged out).
441 if (vma->vm_private_data == NULL) {
442 if (m.addr != vma->vm_start ||
443 m.addr + (nr_pages << PAGE_SHIFT) != vma->vm_end) {
447 if (xen_feature(XENFEAT_auto_translated_physmap)) {
448 ret = alloc_empty_pages(vma, m.num);
452 vma->vm_private_data = PRIV_VMA_LOCKED;
454 if (m.addr < vma->vm_start ||
455 m.addr + (nr_pages << PAGE_SHIFT) > vma->vm_end) {
459 if (privcmd_vma_range_is_mapped(vma, m.addr, nr_pages)) {
465 state.domain = m.dom;
469 state.global_error = 0;
470 state.version = version;
472 /* mmap_batch_fn guarantees ret == 0 */
473 BUG_ON(traverse_pages(m.num, sizeof(xen_pfn_t),
474 &pagelist, mmap_batch_fn, &state));
476 up_write(&mm->mmap_sem);
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);
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))
493 free_page_list(&pagelist);
497 up_write(&mm->mmap_sem);
501 static long privcmd_ioctl(struct file *file,
502 unsigned int cmd, unsigned long data)
505 void __user *udata = (void __user *) data;
508 case IOCTL_PRIVCMD_HYPERCALL:
509 ret = privcmd_ioctl_hypercall(udata);
512 case IOCTL_PRIVCMD_MMAP:
513 ret = privcmd_ioctl_mmap(udata);
516 case IOCTL_PRIVCMD_MMAPBATCH:
517 ret = privcmd_ioctl_mmap_batch(udata, 1);
520 case IOCTL_PRIVCMD_MMAPBATCH_V2:
521 ret = privcmd_ioctl_mmap_batch(udata, 2);
532 static void privcmd_close(struct vm_area_struct *vma)
534 struct page **pages = vma->vm_private_data;
535 int numpgs = (vma->vm_end - vma->vm_start) >> PAGE_SHIFT;
538 if (!xen_feature(XENFEAT_auto_translated_physmap) || !numpgs || !pages)
541 rc = xen_unmap_domain_mfn_range(vma, numpgs, pages);
543 free_xenballooned_pages(numpgs, pages);
545 pr_crit("unable to unmap MFN range: leaking %d pages. rc=%d\n",
550 static int privcmd_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
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);
556 return VM_FAULT_SIGBUS;
559 static struct vm_operations_struct privcmd_vm_ops = {
560 .close = privcmd_close,
561 .fault = privcmd_fault
564 static int privcmd_mmap(struct file *file, struct vm_area_struct *vma)
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;
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.
581 static int is_mapped_fn(pte_t *pte, struct page *pmd_page,
582 unsigned long addr, void *data)
584 return pte_none(*pte) ? 0 : -EBUSY;
587 static int privcmd_vma_range_is_mapped(
588 struct vm_area_struct *vma,
590 unsigned long nr_pages)
592 return apply_to_page_range(vma->vm_mm, addr, nr_pages << PAGE_SHIFT,
593 is_mapped_fn, NULL) != 0;
596 const struct file_operations xen_privcmd_fops = {
597 .owner = THIS_MODULE,
598 .unlocked_ioctl = privcmd_ioctl,
599 .mmap = privcmd_mmap,
601 EXPORT_SYMBOL_GPL(xen_privcmd_fops);
603 static struct miscdevice privcmd_dev = {
604 .minor = MISC_DYNAMIC_MINOR,
605 .name = "xen/privcmd",
606 .fops = &xen_privcmd_fops,
609 static int __init privcmd_init(void)
616 err = misc_register(&privcmd_dev);
618 pr_err("Could not register Xen privcmd device\n");
624 static void __exit privcmd_exit(void)
626 misc_deregister(&privcmd_dev);
629 module_init(privcmd_init);
630 module_exit(privcmd_exit);