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);
162 struct mmap_mfn_state {
164 struct vm_area_struct *vma;
168 static int mmap_mfn_range(void *data, void *state)
170 struct privcmd_mmap_entry *msg = data;
171 struct mmap_mfn_state *st = state;
172 struct vm_area_struct *vma = st->vma;
175 /* Do not allow range to wrap the address space. */
176 if ((msg->npages > (LONG_MAX >> PAGE_SHIFT)) ||
177 ((unsigned long)(msg->npages << PAGE_SHIFT) >= -st->va))
180 /* Range chunks must be contiguous in va space. */
181 if ((msg->va != st->va) ||
182 ((msg->va+(msg->npages<<PAGE_SHIFT)) > vma->vm_end))
185 rc = xen_remap_domain_mfn_range(vma,
187 msg->mfn, msg->npages,
193 st->va += msg->npages << PAGE_SHIFT;
198 static long privcmd_ioctl_mmap(void __user *udata)
200 struct privcmd_mmap mmapcmd;
201 struct mm_struct *mm = current->mm;
202 struct vm_area_struct *vma;
205 struct mmap_mfn_state state;
207 /* We only support privcmd_ioctl_mmap_batch for auto translated. */
208 if (xen_feature(XENFEAT_auto_translated_physmap))
211 if (copy_from_user(&mmapcmd, udata, sizeof(mmapcmd)))
214 rc = gather_array(&pagelist,
215 mmapcmd.num, sizeof(struct privcmd_mmap_entry),
218 if (rc || list_empty(&pagelist))
221 down_write(&mm->mmap_sem);
224 struct page *page = list_first_entry(&pagelist,
226 struct privcmd_mmap_entry *msg = page_address(page);
228 vma = find_vma(mm, msg->va);
231 if (!vma || (msg->va != vma->vm_start) || vma->vm_private_data)
233 vma->vm_private_data = PRIV_VMA_LOCKED;
236 state.va = vma->vm_start;
238 state.domain = mmapcmd.dom;
240 rc = traverse_pages(mmapcmd.num, sizeof(struct privcmd_mmap_entry),
242 mmap_mfn_range, &state);
246 up_write(&mm->mmap_sem);
249 free_page_list(&pagelist);
254 struct mmap_batch_state {
257 struct vm_area_struct *vma;
261 * 1 if at least one error has happened (and no
262 * -ENOENT errors have happened)
263 * -ENOENT if at least 1 -ENOENT has happened.
268 /* User-space mfn array to store errors in the second pass for V1. */
269 xen_pfn_t __user *user_mfn;
270 /* User-space int array to store errors in the second pass for V2. */
271 int __user *user_err;
274 /* auto translated dom0 note: if domU being created is PV, then mfn is
275 * mfn(addr on bus). If it's auto xlated, then mfn is pfn (input to HAP).
277 static int mmap_batch_fn(void *data, void *state)
279 xen_pfn_t *mfnp = data;
280 struct mmap_batch_state *st = state;
281 struct vm_area_struct *vma = st->vma;
282 struct page **pages = vma->vm_private_data;
283 struct page *cur_page = NULL;
286 if (xen_feature(XENFEAT_auto_translated_physmap))
287 cur_page = pages[st->index++];
289 ret = xen_remap_domain_mfn_range(st->vma, st->va & PAGE_MASK, *mfnp, 1,
290 st->vma->vm_page_prot, st->domain,
293 /* Store error code for second pass. */
294 if (st->version == 1) {
297 * V1 encodes the error codes in the 32bit top nibble of the
298 * mfn (with its known limitations vis-a-vis 64 bit callers).
300 *mfnp |= (ret == -ENOENT) ?
301 PRIVCMD_MMAPBATCH_PAGED_ERROR :
302 PRIVCMD_MMAPBATCH_MFN_ERROR;
304 } else { /* st->version == 2 */
305 *((int *) mfnp) = ret;
308 /* And see if it affects the global_error. */
311 st->global_error = -ENOENT;
313 /* Record that at least one error has happened. */
314 if (st->global_error == 0)
315 st->global_error = 1;
323 static int mmap_return_errors(void *data, void *state)
325 struct mmap_batch_state *st = state;
327 if (st->version == 1) {
328 xen_pfn_t mfnp = *((xen_pfn_t *) data);
329 if (mfnp & PRIVCMD_MMAPBATCH_MFN_ERROR)
330 return __put_user(mfnp, st->user_mfn++);
333 } else { /* st->version == 2 */
334 int err = *((int *) data);
336 return __put_user(err, st->user_err++);
344 /* Allocate pfns that are then mapped with gmfns from foreign domid. Update
345 * the vma with the page info to use later.
346 * Returns: 0 if success, otherwise -errno
348 static int alloc_empty_pages(struct vm_area_struct *vma, int numpgs)
353 pages = kcalloc(numpgs, sizeof(pages[0]), GFP_KERNEL);
357 rc = alloc_xenballooned_pages(numpgs, pages, 0);
359 pr_warn("%s Could not alloc %d pfns rc:%d\n", __func__,
364 BUG_ON(vma->vm_private_data != NULL);
365 vma->vm_private_data = pages;
370 static struct vm_operations_struct privcmd_vm_ops;
372 static long privcmd_ioctl_mmap_batch(void __user *udata, int version)
375 struct privcmd_mmapbatch_v2 m;
376 struct mm_struct *mm = current->mm;
377 struct vm_area_struct *vma;
378 unsigned long nr_pages;
380 struct mmap_batch_state state;
384 if (copy_from_user(&m, udata, sizeof(struct privcmd_mmapbatch)))
386 /* Returns per-frame error in m.arr. */
388 if (!access_ok(VERIFY_WRITE, m.arr, m.num * sizeof(*m.arr)))
392 if (copy_from_user(&m, udata, sizeof(struct privcmd_mmapbatch_v2)))
394 /* Returns per-frame error code in m.err. */
395 if (!access_ok(VERIFY_WRITE, m.err, m.num * (sizeof(*m.err))))
403 if ((m.num <= 0) || (nr_pages > (LONG_MAX >> PAGE_SHIFT)))
406 ret = gather_array(&pagelist, m.num, sizeof(xen_pfn_t), m.arr);
410 if (list_empty(&pagelist)) {
416 /* Zero error array now to only copy back actual errors. */
417 if (clear_user(m.err, sizeof(int) * m.num)) {
423 down_write(&mm->mmap_sem);
425 vma = find_vma(mm, m.addr);
427 vma->vm_ops != &privcmd_vm_ops) {
433 * Caller must either:
435 * Map the whole VMA range, which will also allocate all the
436 * pages required for the auto_translated_physmap case.
440 * Map unmapped holes left from a previous map attempt (e.g.,
441 * because those foreign frames were previously paged out).
443 if (vma->vm_private_data == NULL) {
444 if (m.addr != vma->vm_start ||
445 m.addr + (nr_pages << PAGE_SHIFT) != vma->vm_end) {
449 if (xen_feature(XENFEAT_auto_translated_physmap)) {
450 ret = alloc_empty_pages(vma, m.num);
454 vma->vm_private_data = PRIV_VMA_LOCKED;
456 if (m.addr < vma->vm_start ||
457 m.addr + (nr_pages << PAGE_SHIFT) > vma->vm_end) {
461 if (privcmd_vma_range_is_mapped(vma, m.addr, nr_pages)) {
467 state.domain = m.dom;
471 state.global_error = 0;
472 state.version = version;
474 /* mmap_batch_fn guarantees ret == 0 */
475 BUG_ON(traverse_pages(m.num, sizeof(xen_pfn_t),
476 &pagelist, mmap_batch_fn, &state));
478 up_write(&mm->mmap_sem);
480 if (state.global_error) {
481 /* Write back errors in second pass. */
482 state.user_mfn = (xen_pfn_t *)m.arr;
483 state.user_err = m.err;
484 ret = traverse_pages(m.num, sizeof(xen_pfn_t),
485 &pagelist, mmap_return_errors, &state);
489 /* If we have not had any EFAULT-like global errors then set the global
490 * error to -ENOENT if necessary. */
491 if ((ret == 0) && (state.global_error == -ENOENT))
495 free_page_list(&pagelist);
499 up_write(&mm->mmap_sem);
503 static long privcmd_ioctl(struct file *file,
504 unsigned int cmd, unsigned long data)
507 void __user *udata = (void __user *) data;
510 case IOCTL_PRIVCMD_HYPERCALL:
511 ret = privcmd_ioctl_hypercall(udata);
514 case IOCTL_PRIVCMD_MMAP:
515 ret = privcmd_ioctl_mmap(udata);
518 case IOCTL_PRIVCMD_MMAPBATCH:
519 ret = privcmd_ioctl_mmap_batch(udata, 1);
522 case IOCTL_PRIVCMD_MMAPBATCH_V2:
523 ret = privcmd_ioctl_mmap_batch(udata, 2);
534 static void privcmd_close(struct vm_area_struct *vma)
536 struct page **pages = vma->vm_private_data;
537 int numpgs = (vma->vm_end - vma->vm_start) >> PAGE_SHIFT;
540 if (!xen_feature(XENFEAT_auto_translated_physmap) || !numpgs || !pages)
543 rc = xen_unmap_domain_mfn_range(vma, numpgs, pages);
545 free_xenballooned_pages(numpgs, pages);
547 pr_crit("unable to unmap MFN range: leaking %d pages. rc=%d\n",
552 static int privcmd_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
554 printk(KERN_DEBUG "privcmd_fault: vma=%p %lx-%lx, pgoff=%lx, uv=%p\n",
555 vma, vma->vm_start, vma->vm_end,
556 vmf->pgoff, vmf->virtual_address);
558 return VM_FAULT_SIGBUS;
561 static struct vm_operations_struct privcmd_vm_ops = {
562 .close = privcmd_close,
563 .fault = privcmd_fault
566 static int privcmd_mmap(struct file *file, struct vm_area_struct *vma)
568 /* DONTCOPY is essential for Xen because copy_page_range doesn't know
569 * how to recreate these mappings */
570 vma->vm_flags |= VM_IO | VM_PFNMAP | VM_DONTCOPY |
571 VM_DONTEXPAND | VM_DONTDUMP;
572 vma->vm_ops = &privcmd_vm_ops;
573 vma->vm_private_data = NULL;
579 * For MMAPBATCH*. This allows asserting the singleshot mapping
580 * on a per pfn/pte basis. Mapping calls that fail with ENOENT
581 * can be then retried until success.
583 static int is_mapped_fn(pte_t *pte, struct page *pmd_page,
584 unsigned long addr, void *data)
586 return pte_none(*pte) ? 0 : -EBUSY;
589 static int privcmd_vma_range_is_mapped(
590 struct vm_area_struct *vma,
592 unsigned long nr_pages)
594 return apply_to_page_range(vma->vm_mm, addr, nr_pages << PAGE_SHIFT,
595 is_mapped_fn, NULL) != 0;
598 const struct file_operations xen_privcmd_fops = {
599 .owner = THIS_MODULE,
600 .unlocked_ioctl = privcmd_ioctl,
601 .mmap = privcmd_mmap,
603 EXPORT_SYMBOL_GPL(xen_privcmd_fops);
605 static struct miscdevice privcmd_dev = {
606 .minor = MISC_DYNAMIC_MINOR,
607 .name = "xen/privcmd",
608 .fops = &xen_privcmd_fops,
611 static int __init privcmd_init(void)
618 err = misc_register(&privcmd_dev);
620 pr_err("Could not register Xen privcmd device\n");
626 static void __exit privcmd_exit(void)
628 misc_deregister(&privcmd_dev);
631 module_init(privcmd_init);
632 module_exit(privcmd_exit);