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 #ifndef HAVE_ARCH_PRIVCMD_MMAP
47 static int privcmd_enforce_singleshot_mapping(struct vm_area_struct *vma);
50 static long privcmd_ioctl_hypercall(void __user *udata)
52 struct privcmd_hypercall hypercall;
55 if (copy_from_user(&hypercall, udata, sizeof(hypercall)))
58 ret = privcmd_call(hypercall.op,
59 hypercall.arg[0], hypercall.arg[1],
60 hypercall.arg[2], hypercall.arg[3],
66 static void free_page_list(struct list_head *pages)
70 list_for_each_entry_safe(p, n, pages, lru)
73 INIT_LIST_HEAD(pages);
77 * Given an array of items in userspace, return a list of pages
78 * containing the data. If copying fails, either because of memory
79 * allocation failure or a problem reading user memory, return an
80 * error code; its up to the caller to dispose of any partial list.
82 static int gather_array(struct list_head *pagelist,
83 unsigned nelem, size_t size,
84 const void __user *data)
94 pagedata = NULL; /* quiet, gcc */
96 if (pageidx > PAGE_SIZE-size) {
97 struct page *page = alloc_page(GFP_KERNEL);
103 pagedata = page_address(page);
105 list_add_tail(&page->lru, pagelist);
110 if (copy_from_user(pagedata + pageidx, data, size))
124 * Call function "fn" on each element of the array fragmented
125 * over a list of pages.
127 static int traverse_pages(unsigned nelem, size_t size,
128 struct list_head *pos,
129 int (*fn)(void *data, void *state),
136 BUG_ON(size > PAGE_SIZE);
139 pagedata = NULL; /* hush, gcc */
142 if (pageidx > PAGE_SIZE-size) {
145 page = list_entry(pos, struct page, lru);
146 pagedata = page_address(page);
150 ret = (*fn)(pagedata + pageidx, state);
159 struct mmap_mfn_state {
161 struct vm_area_struct *vma;
165 static int mmap_mfn_range(void *data, void *state)
167 struct privcmd_mmap_entry *msg = data;
168 struct mmap_mfn_state *st = state;
169 struct vm_area_struct *vma = st->vma;
172 /* Do not allow range to wrap the address space. */
173 if ((msg->npages > (LONG_MAX >> PAGE_SHIFT)) ||
174 ((unsigned long)(msg->npages << PAGE_SHIFT) >= -st->va))
177 /* Range chunks must be contiguous in va space. */
178 if ((msg->va != st->va) ||
179 ((msg->va+(msg->npages<<PAGE_SHIFT)) > vma->vm_end))
182 rc = xen_remap_domain_mfn_range(vma,
184 msg->mfn, msg->npages,
190 st->va += msg->npages << PAGE_SHIFT;
195 static long privcmd_ioctl_mmap(void __user *udata)
197 struct privcmd_mmap mmapcmd;
198 struct mm_struct *mm = current->mm;
199 struct vm_area_struct *vma;
202 struct mmap_mfn_state state;
204 /* We only support privcmd_ioctl_mmap_batch for auto translated. */
205 if (xen_feature(XENFEAT_auto_translated_physmap))
208 if (copy_from_user(&mmapcmd, udata, sizeof(mmapcmd)))
211 rc = gather_array(&pagelist,
212 mmapcmd.num, sizeof(struct privcmd_mmap_entry),
215 if (rc || list_empty(&pagelist))
218 down_write(&mm->mmap_sem);
221 struct page *page = list_first_entry(&pagelist,
223 struct privcmd_mmap_entry *msg = page_address(page);
225 vma = find_vma(mm, msg->va);
228 if (!vma || (msg->va != vma->vm_start) ||
229 !privcmd_enforce_singleshot_mapping(vma))
233 state.va = vma->vm_start;
235 state.domain = mmapcmd.dom;
237 rc = traverse_pages(mmapcmd.num, sizeof(struct privcmd_mmap_entry),
239 mmap_mfn_range, &state);
243 up_write(&mm->mmap_sem);
246 free_page_list(&pagelist);
251 struct mmap_batch_state {
254 struct vm_area_struct *vma;
258 * 1 if at least one error has happened (and no
259 * -ENOENT errors have happened)
260 * -ENOENT if at least 1 -ENOENT has happened.
265 /* User-space mfn array to store errors in the second pass for V1. */
266 xen_pfn_t __user *user_mfn;
267 /* User-space int array to store errors in the second pass for V2. */
268 int __user *user_err;
271 /* auto translated dom0 note: if domU being created is PV, then mfn is
272 * mfn(addr on bus). If it's auto xlated, then mfn is pfn (input to HAP).
274 static int mmap_batch_fn(void *data, void *state)
276 xen_pfn_t *mfnp = data;
277 struct mmap_batch_state *st = state;
278 struct vm_area_struct *vma = st->vma;
279 struct page **pages = vma->vm_private_data;
280 struct page *cur_page = NULL;
283 if (xen_feature(XENFEAT_auto_translated_physmap))
284 cur_page = pages[st->index++];
286 ret = xen_remap_domain_mfn_range(st->vma, st->va & PAGE_MASK, *mfnp, 1,
287 st->vma->vm_page_prot, st->domain,
290 /* Store error code for second pass. */
291 if (st->version == 1) {
294 * V1 encodes the error codes in the 32bit top nibble of the
295 * mfn (with its known limitations vis-a-vis 64 bit callers).
297 *mfnp |= (ret == -ENOENT) ?
298 PRIVCMD_MMAPBATCH_PAGED_ERROR :
299 PRIVCMD_MMAPBATCH_MFN_ERROR;
301 } else { /* st->version == 2 */
302 *((int *) mfnp) = ret;
305 /* And see if it affects the global_error. */
308 st->global_error = -ENOENT;
310 /* Record that at least one error has happened. */
311 if (st->global_error == 0)
312 st->global_error = 1;
320 static int mmap_return_errors(void *data, void *state)
322 struct mmap_batch_state *st = state;
324 if (st->version == 1) {
325 xen_pfn_t mfnp = *((xen_pfn_t *) data);
326 if (mfnp & PRIVCMD_MMAPBATCH_MFN_ERROR)
327 return __put_user(mfnp, st->user_mfn++);
330 } else { /* st->version == 2 */
331 int err = *((int *) data);
333 return __put_user(err, st->user_err++);
341 /* Allocate pfns that are then mapped with gmfns from foreign domid. Update
342 * the vma with the page info to use later.
343 * Returns: 0 if success, otherwise -errno
345 static int alloc_empty_pages(struct vm_area_struct *vma, int numpgs)
350 pages = kcalloc(numpgs, sizeof(pages[0]), GFP_KERNEL);
354 rc = alloc_xenballooned_pages(numpgs, pages, 0);
356 pr_warn("%s Could not alloc %d pfns rc:%d\n", __func__,
361 BUG_ON(vma->vm_private_data != PRIV_VMA_LOCKED);
362 vma->vm_private_data = pages;
367 static struct vm_operations_struct privcmd_vm_ops;
369 static long privcmd_ioctl_mmap_batch(void __user *udata, int version)
372 struct privcmd_mmapbatch_v2 m;
373 struct mm_struct *mm = current->mm;
374 struct vm_area_struct *vma;
375 unsigned long nr_pages;
377 struct mmap_batch_state state;
381 if (copy_from_user(&m, udata, sizeof(struct privcmd_mmapbatch)))
383 /* Returns per-frame error in m.arr. */
385 if (!access_ok(VERIFY_WRITE, m.arr, m.num * sizeof(*m.arr)))
389 if (copy_from_user(&m, udata, sizeof(struct privcmd_mmapbatch_v2)))
391 /* Returns per-frame error code in m.err. */
392 if (!access_ok(VERIFY_WRITE, m.err, m.num * (sizeof(*m.err))))
400 if ((m.num <= 0) || (nr_pages > (LONG_MAX >> PAGE_SHIFT)))
403 ret = gather_array(&pagelist, m.num, sizeof(xen_pfn_t), m.arr);
407 if (list_empty(&pagelist)) {
413 /* Zero error array now to only copy back actual errors. */
414 if (clear_user(m.err, sizeof(int) * m.num)) {
420 down_write(&mm->mmap_sem);
422 vma = find_vma(mm, m.addr);
424 vma->vm_ops != &privcmd_vm_ops ||
425 (m.addr != vma->vm_start) ||
426 ((m.addr + (nr_pages << PAGE_SHIFT)) != vma->vm_end) ||
427 !privcmd_enforce_singleshot_mapping(vma)) {
428 up_write(&mm->mmap_sem);
432 if (xen_feature(XENFEAT_auto_translated_physmap)) {
433 ret = alloc_empty_pages(vma, m.num);
435 up_write(&mm->mmap_sem);
440 state.domain = m.dom;
444 state.global_error = 0;
445 state.version = version;
447 /* mmap_batch_fn guarantees ret == 0 */
448 BUG_ON(traverse_pages(m.num, sizeof(xen_pfn_t),
449 &pagelist, mmap_batch_fn, &state));
451 up_write(&mm->mmap_sem);
453 if (state.global_error) {
454 /* Write back errors in second pass. */
455 state.user_mfn = (xen_pfn_t *)m.arr;
456 state.user_err = m.err;
457 ret = traverse_pages(m.num, sizeof(xen_pfn_t),
458 &pagelist, mmap_return_errors, &state);
462 /* If we have not had any EFAULT-like global errors then set the global
463 * error to -ENOENT if necessary. */
464 if ((ret == 0) && (state.global_error == -ENOENT))
468 free_page_list(&pagelist);
473 static long privcmd_ioctl(struct file *file,
474 unsigned int cmd, unsigned long data)
477 void __user *udata = (void __user *) data;
480 case IOCTL_PRIVCMD_HYPERCALL:
481 ret = privcmd_ioctl_hypercall(udata);
484 case IOCTL_PRIVCMD_MMAP:
485 ret = privcmd_ioctl_mmap(udata);
488 case IOCTL_PRIVCMD_MMAPBATCH:
489 ret = privcmd_ioctl_mmap_batch(udata, 1);
492 case IOCTL_PRIVCMD_MMAPBATCH_V2:
493 ret = privcmd_ioctl_mmap_batch(udata, 2);
504 static void privcmd_close(struct vm_area_struct *vma)
506 struct page **pages = vma->vm_private_data;
507 int numpgs = (vma->vm_end - vma->vm_start) >> PAGE_SHIFT;
509 if (!xen_feature(XENFEAT_auto_translated_physmap) || !numpgs || !pages)
512 xen_unmap_domain_mfn_range(vma, numpgs, pages);
513 free_xenballooned_pages(numpgs, pages);
517 static int privcmd_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
519 printk(KERN_DEBUG "privcmd_fault: vma=%p %lx-%lx, pgoff=%lx, uv=%p\n",
520 vma, vma->vm_start, vma->vm_end,
521 vmf->pgoff, vmf->virtual_address);
523 return VM_FAULT_SIGBUS;
526 static struct vm_operations_struct privcmd_vm_ops = {
527 .close = privcmd_close,
528 .fault = privcmd_fault
531 static int privcmd_mmap(struct file *file, struct vm_area_struct *vma)
533 /* DONTCOPY is essential for Xen because copy_page_range doesn't know
534 * how to recreate these mappings */
535 vma->vm_flags |= VM_IO | VM_PFNMAP | VM_DONTCOPY |
536 VM_DONTEXPAND | VM_DONTDUMP;
537 vma->vm_ops = &privcmd_vm_ops;
538 vma->vm_private_data = NULL;
543 static int privcmd_enforce_singleshot_mapping(struct vm_area_struct *vma)
545 return !cmpxchg(&vma->vm_private_data, NULL, PRIV_VMA_LOCKED);
548 const struct file_operations xen_privcmd_fops = {
549 .owner = THIS_MODULE,
550 .unlocked_ioctl = privcmd_ioctl,
551 .mmap = privcmd_mmap,
553 EXPORT_SYMBOL_GPL(xen_privcmd_fops);
555 static struct miscdevice privcmd_dev = {
556 .minor = MISC_DYNAMIC_MINOR,
557 .name = "xen/privcmd",
558 .fops = &xen_privcmd_fops,
561 static int __init privcmd_init(void)
568 err = misc_register(&privcmd_dev);
570 pr_err("Could not register Xen privcmd device\n");
576 static void __exit privcmd_exit(void)
578 misc_deregister(&privcmd_dev);
581 module_init(privcmd_init);
582 module_exit(privcmd_exit);