1 /******************************************************************************
4 * Interface to privileged domain-0 commands.
6 * Copyright (c) 2002-2004, K A Fraser, B Dragovic
9 #include <linux/kernel.h>
10 #include <linux/module.h>
11 #include <linux/sched.h>
12 #include <linux/slab.h>
13 #include <linux/string.h>
14 #include <linux/errno.h>
16 #include <linux/mman.h>
17 #include <linux/uaccess.h>
18 #include <linux/swap.h>
19 #include <linux/highmem.h>
20 #include <linux/pagemap.h>
21 #include <linux/seq_file.h>
22 #include <linux/miscdevice.h>
24 #include <asm/pgalloc.h>
25 #include <asm/pgtable.h>
27 #include <asm/xen/hypervisor.h>
28 #include <asm/xen/hypercall.h>
31 #include <xen/privcmd.h>
32 #include <xen/interface/xen.h>
33 #include <xen/features.h>
35 #include <xen/xen-ops.h>
36 #include <xen/balloon.h>
40 MODULE_LICENSE("GPL");
42 #define PRIV_VMA_LOCKED ((void *)1)
44 #ifndef HAVE_ARCH_PRIVCMD_MMAP
45 static int privcmd_enforce_singleshot_mapping(struct vm_area_struct *vma);
48 static long privcmd_ioctl_hypercall(void __user *udata)
50 struct privcmd_hypercall hypercall;
53 if (copy_from_user(&hypercall, udata, sizeof(hypercall)))
56 ret = privcmd_call(hypercall.op,
57 hypercall.arg[0], hypercall.arg[1],
58 hypercall.arg[2], hypercall.arg[3],
64 static void free_page_list(struct list_head *pages)
68 list_for_each_entry_safe(p, n, pages, lru)
71 INIT_LIST_HEAD(pages);
75 * Given an array of items in userspace, return a list of pages
76 * containing the data. If copying fails, either because of memory
77 * allocation failure or a problem reading user memory, return an
78 * error code; its up to the caller to dispose of any partial list.
80 static int gather_array(struct list_head *pagelist,
81 unsigned nelem, size_t size,
82 const void __user *data)
92 pagedata = NULL; /* quiet, gcc */
94 if (pageidx > PAGE_SIZE-size) {
95 struct page *page = alloc_page(GFP_KERNEL);
101 pagedata = page_address(page);
103 list_add_tail(&page->lru, pagelist);
108 if (copy_from_user(pagedata + pageidx, data, size))
122 * Call function "fn" on each element of the array fragmented
123 * over a list of pages.
125 static int traverse_pages(unsigned nelem, size_t size,
126 struct list_head *pos,
127 int (*fn)(void *data, void *state),
134 BUG_ON(size > PAGE_SIZE);
137 pagedata = NULL; /* hush, gcc */
140 if (pageidx > PAGE_SIZE-size) {
143 page = list_entry(pos, struct page, lru);
144 pagedata = page_address(page);
148 ret = (*fn)(pagedata + pageidx, state);
157 struct mmap_mfn_state {
159 struct vm_area_struct *vma;
163 static int mmap_mfn_range(void *data, void *state)
165 struct privcmd_mmap_entry *msg = data;
166 struct mmap_mfn_state *st = state;
167 struct vm_area_struct *vma = st->vma;
170 /* Do not allow range to wrap the address space. */
171 if ((msg->npages > (LONG_MAX >> PAGE_SHIFT)) ||
172 ((unsigned long)(msg->npages << PAGE_SHIFT) >= -st->va))
175 /* Range chunks must be contiguous in va space. */
176 if ((msg->va != st->va) ||
177 ((msg->va+(msg->npages<<PAGE_SHIFT)) > vma->vm_end))
180 rc = xen_remap_domain_mfn_range(vma,
182 msg->mfn, msg->npages,
188 st->va += msg->npages << PAGE_SHIFT;
193 static long privcmd_ioctl_mmap(void __user *udata)
195 struct privcmd_mmap mmapcmd;
196 struct mm_struct *mm = current->mm;
197 struct vm_area_struct *vma;
200 struct mmap_mfn_state state;
202 /* We only support privcmd_ioctl_mmap_batch for auto translated. */
203 if (xen_feature(XENFEAT_auto_translated_physmap))
206 if (copy_from_user(&mmapcmd, udata, sizeof(mmapcmd)))
209 rc = gather_array(&pagelist,
210 mmapcmd.num, sizeof(struct privcmd_mmap_entry),
213 if (rc || list_empty(&pagelist))
216 down_write(&mm->mmap_sem);
219 struct page *page = list_first_entry(&pagelist,
221 struct privcmd_mmap_entry *msg = page_address(page);
223 vma = find_vma(mm, msg->va);
226 if (!vma || (msg->va != vma->vm_start) ||
227 !privcmd_enforce_singleshot_mapping(vma))
231 state.va = vma->vm_start;
233 state.domain = mmapcmd.dom;
235 rc = traverse_pages(mmapcmd.num, sizeof(struct privcmd_mmap_entry),
237 mmap_mfn_range, &state);
241 up_write(&mm->mmap_sem);
244 free_page_list(&pagelist);
249 struct mmap_batch_state {
252 struct vm_area_struct *vma;
256 * 1 if at least one error has happened (and no
257 * -ENOENT errors have happened)
258 * -ENOENT if at least 1 -ENOENT has happened.
261 /* An array for individual errors */
264 /* User-space mfn array to store errors in the second pass for V1. */
265 xen_pfn_t __user *user_mfn;
268 /* auto translated dom0 note: if domU being created is PV, then mfn is
269 * mfn(addr on bus). If it's auto xlated, then mfn is pfn (input to HAP).
271 static int mmap_batch_fn(void *data, void *state)
273 xen_pfn_t *mfnp = data;
274 struct mmap_batch_state *st = state;
275 struct vm_area_struct *vma = st->vma;
276 struct page **pages = vma->vm_private_data;
277 struct page *cur_page = NULL;
280 if (xen_feature(XENFEAT_auto_translated_physmap))
281 cur_page = pages[st->index++];
283 ret = xen_remap_domain_mfn_range(st->vma, st->va & PAGE_MASK, *mfnp, 1,
284 st->vma->vm_page_prot, st->domain,
287 /* Store error code for second pass. */
290 /* And see if it affects the global_error. */
293 st->global_error = -ENOENT;
295 /* Record that at least one error has happened. */
296 if (st->global_error == 0)
297 st->global_error = 1;
305 static int mmap_return_errors_v1(void *data, void *state)
307 xen_pfn_t *mfnp = data;
308 struct mmap_batch_state *st = state;
309 int err = *(st->err++);
312 * V1 encodes the error codes in the 32bit top nibble of the
313 * mfn (with its known limitations vis-a-vis 64 bit callers).
315 *mfnp |= (err == -ENOENT) ?
316 PRIVCMD_MMAPBATCH_PAGED_ERROR :
317 PRIVCMD_MMAPBATCH_MFN_ERROR;
318 return __put_user(*mfnp, st->user_mfn++);
321 /* Allocate pfns that are then mapped with gmfns from foreign domid. Update
322 * the vma with the page info to use later.
323 * Returns: 0 if success, otherwise -errno
325 static int alloc_empty_pages(struct vm_area_struct *vma, int numpgs)
330 pages = kcalloc(numpgs, sizeof(pages[0]), GFP_KERNEL);
334 rc = alloc_xenballooned_pages(numpgs, pages, 0);
336 pr_warn("%s Could not alloc %d pfns rc:%d\n", __func__,
341 BUG_ON(vma->vm_private_data != PRIV_VMA_LOCKED);
342 vma->vm_private_data = pages;
347 static struct vm_operations_struct privcmd_vm_ops;
349 static long privcmd_ioctl_mmap_batch(void __user *udata, int version)
352 struct privcmd_mmapbatch_v2 m;
353 struct mm_struct *mm = current->mm;
354 struct vm_area_struct *vma;
355 unsigned long nr_pages;
357 int *err_array = NULL;
358 struct mmap_batch_state state;
362 if (copy_from_user(&m, udata, sizeof(struct privcmd_mmapbatch)))
364 /* Returns per-frame error in m.arr. */
366 if (!access_ok(VERIFY_WRITE, m.arr, m.num * sizeof(*m.arr)))
370 if (copy_from_user(&m, udata, sizeof(struct privcmd_mmapbatch_v2)))
372 /* Returns per-frame error code in m.err. */
373 if (!access_ok(VERIFY_WRITE, m.err, m.num * (sizeof(*m.err))))
381 if ((m.num <= 0) || (nr_pages > (LONG_MAX >> PAGE_SHIFT)))
384 ret = gather_array(&pagelist, m.num, sizeof(xen_pfn_t), m.arr);
388 if (list_empty(&pagelist)) {
393 err_array = kcalloc(m.num, sizeof(int), GFP_KERNEL);
394 if (err_array == NULL) {
399 down_write(&mm->mmap_sem);
401 vma = find_vma(mm, m.addr);
403 vma->vm_ops != &privcmd_vm_ops ||
404 (m.addr != vma->vm_start) ||
405 ((m.addr + (nr_pages << PAGE_SHIFT)) != vma->vm_end) ||
406 !privcmd_enforce_singleshot_mapping(vma)) {
407 up_write(&mm->mmap_sem);
411 if (xen_feature(XENFEAT_auto_translated_physmap)) {
412 ret = alloc_empty_pages(vma, m.num);
414 up_write(&mm->mmap_sem);
419 state.domain = m.dom;
423 state.global_error = 0;
424 state.err = err_array;
426 /* mmap_batch_fn guarantees ret == 0 */
427 BUG_ON(traverse_pages(m.num, sizeof(xen_pfn_t),
428 &pagelist, mmap_batch_fn, &state));
430 up_write(&mm->mmap_sem);
433 if (state.global_error) {
434 /* Write back errors in second pass. */
435 state.user_mfn = (xen_pfn_t *)m.arr;
436 state.err = err_array;
437 ret = traverse_pages(m.num, sizeof(xen_pfn_t),
438 &pagelist, mmap_return_errors_v1, &state);
442 } else if (version == 2) {
443 ret = __copy_to_user(m.err, err_array, m.num * sizeof(int));
448 /* If we have not had any EFAULT-like global errors then set the global
449 * error to -ENOENT if necessary. */
450 if ((ret == 0) && (state.global_error == -ENOENT))
455 free_page_list(&pagelist);
460 static long privcmd_ioctl(struct file *file,
461 unsigned int cmd, unsigned long data)
464 void __user *udata = (void __user *) data;
467 case IOCTL_PRIVCMD_HYPERCALL:
468 ret = privcmd_ioctl_hypercall(udata);
471 case IOCTL_PRIVCMD_MMAP:
472 ret = privcmd_ioctl_mmap(udata);
475 case IOCTL_PRIVCMD_MMAPBATCH:
476 ret = privcmd_ioctl_mmap_batch(udata, 1);
479 case IOCTL_PRIVCMD_MMAPBATCH_V2:
480 ret = privcmd_ioctl_mmap_batch(udata, 2);
491 static void privcmd_close(struct vm_area_struct *vma)
493 struct page **pages = vma->vm_private_data;
494 int numpgs = (vma->vm_end - vma->vm_start) >> PAGE_SHIFT;
496 if (!xen_feature(XENFEAT_auto_translated_physmap || !numpgs || !pages))
499 xen_unmap_domain_mfn_range(vma, numpgs, pages);
500 free_xenballooned_pages(numpgs, pages);
504 static int privcmd_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
506 printk(KERN_DEBUG "privcmd_fault: vma=%p %lx-%lx, pgoff=%lx, uv=%p\n",
507 vma, vma->vm_start, vma->vm_end,
508 vmf->pgoff, vmf->virtual_address);
510 return VM_FAULT_SIGBUS;
513 static struct vm_operations_struct privcmd_vm_ops = {
514 .close = privcmd_close,
515 .fault = privcmd_fault
518 static int privcmd_mmap(struct file *file, struct vm_area_struct *vma)
520 /* DONTCOPY is essential for Xen because copy_page_range doesn't know
521 * how to recreate these mappings */
522 vma->vm_flags |= VM_IO | VM_PFNMAP | VM_DONTCOPY |
523 VM_DONTEXPAND | VM_DONTDUMP;
524 vma->vm_ops = &privcmd_vm_ops;
525 vma->vm_private_data = NULL;
530 static int privcmd_enforce_singleshot_mapping(struct vm_area_struct *vma)
532 return !cmpxchg(&vma->vm_private_data, NULL, PRIV_VMA_LOCKED);
535 const struct file_operations xen_privcmd_fops = {
536 .owner = THIS_MODULE,
537 .unlocked_ioctl = privcmd_ioctl,
538 .mmap = privcmd_mmap,
540 EXPORT_SYMBOL_GPL(xen_privcmd_fops);
542 static struct miscdevice privcmd_dev = {
543 .minor = MISC_DYNAMIC_MINOR,
544 .name = "xen/privcmd",
545 .fops = &xen_privcmd_fops,
548 static int __init privcmd_init(void)
555 err = misc_register(&privcmd_dev);
557 printk(KERN_ERR "Could not register Xen privcmd device\n");
563 static void __exit privcmd_exit(void)
565 misc_deregister(&privcmd_dev);
568 module_init(privcmd_init);
569 module_exit(privcmd_exit);