Merge git://git.kernel.org/pub/scm/linux/kernel/git/davem/net
[linux-2.6.git] / drivers / xen / privcmd.c
1 /******************************************************************************
2  * privcmd.c
3  *
4  * Interface to privileged domain-0 commands.
5  *
6  * Copyright (c) 2002-2004, K A Fraser, B Dragovic
7  */
8
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>
15 #include <linux/mm.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>
23
24 #include <asm/pgalloc.h>
25 #include <asm/pgtable.h>
26 #include <asm/tlb.h>
27 #include <asm/xen/hypervisor.h>
28 #include <asm/xen/hypercall.h>
29
30 #include <xen/xen.h>
31 #include <xen/privcmd.h>
32 #include <xen/interface/xen.h>
33 #include <xen/features.h>
34 #include <xen/page.h>
35 #include <xen/xen-ops.h>
36
37 #include "privcmd.h"
38
39 MODULE_LICENSE("GPL");
40
41 #ifndef HAVE_ARCH_PRIVCMD_MMAP
42 static int privcmd_enforce_singleshot_mapping(struct vm_area_struct *vma);
43 #endif
44
45 static long privcmd_ioctl_hypercall(void __user *udata)
46 {
47         struct privcmd_hypercall hypercall;
48         long ret;
49
50         if (copy_from_user(&hypercall, udata, sizeof(hypercall)))
51                 return -EFAULT;
52
53         ret = privcmd_call(hypercall.op,
54                            hypercall.arg[0], hypercall.arg[1],
55                            hypercall.arg[2], hypercall.arg[3],
56                            hypercall.arg[4]);
57
58         return ret;
59 }
60
61 static void free_page_list(struct list_head *pages)
62 {
63         struct page *p, *n;
64
65         list_for_each_entry_safe(p, n, pages, lru)
66                 __free_page(p);
67
68         INIT_LIST_HEAD(pages);
69 }
70
71 /*
72  * Given an array of items in userspace, return a list of pages
73  * containing the data.  If copying fails, either because of memory
74  * allocation failure or a problem reading user memory, return an
75  * error code; its up to the caller to dispose of any partial list.
76  */
77 static int gather_array(struct list_head *pagelist,
78                         unsigned nelem, size_t size,
79                         void __user *data)
80 {
81         unsigned pageidx;
82         void *pagedata;
83         int ret;
84
85         if (size > PAGE_SIZE)
86                 return 0;
87
88         pageidx = PAGE_SIZE;
89         pagedata = NULL;        /* quiet, gcc */
90         while (nelem--) {
91                 if (pageidx > PAGE_SIZE-size) {
92                         struct page *page = alloc_page(GFP_KERNEL);
93
94                         ret = -ENOMEM;
95                         if (page == NULL)
96                                 goto fail;
97
98                         pagedata = page_address(page);
99
100                         list_add_tail(&page->lru, pagelist);
101                         pageidx = 0;
102                 }
103
104                 ret = -EFAULT;
105                 if (copy_from_user(pagedata + pageidx, data, size))
106                         goto fail;
107
108                 data += size;
109                 pageidx += size;
110         }
111
112         ret = 0;
113
114 fail:
115         return ret;
116 }
117
118 /*
119  * Call function "fn" on each element of the array fragmented
120  * over a list of pages.
121  */
122 static int traverse_pages(unsigned nelem, size_t size,
123                           struct list_head *pos,
124                           int (*fn)(void *data, void *state),
125                           void *state)
126 {
127         void *pagedata;
128         unsigned pageidx;
129         int ret = 0;
130
131         BUG_ON(size > PAGE_SIZE);
132
133         pageidx = PAGE_SIZE;
134         pagedata = NULL;        /* hush, gcc */
135
136         while (nelem--) {
137                 if (pageidx > PAGE_SIZE-size) {
138                         struct page *page;
139                         pos = pos->next;
140                         page = list_entry(pos, struct page, lru);
141                         pagedata = page_address(page);
142                         pageidx = 0;
143                 }
144
145                 ret = (*fn)(pagedata + pageidx, state);
146                 if (ret)
147                         break;
148                 pageidx += size;
149         }
150
151         return ret;
152 }
153
154 struct mmap_mfn_state {
155         unsigned long va;
156         struct vm_area_struct *vma;
157         domid_t domain;
158 };
159
160 static int mmap_mfn_range(void *data, void *state)
161 {
162         struct privcmd_mmap_entry *msg = data;
163         struct mmap_mfn_state *st = state;
164         struct vm_area_struct *vma = st->vma;
165         int rc;
166
167         /* Do not allow range to wrap the address space. */
168         if ((msg->npages > (LONG_MAX >> PAGE_SHIFT)) ||
169             ((unsigned long)(msg->npages << PAGE_SHIFT) >= -st->va))
170                 return -EINVAL;
171
172         /* Range chunks must be contiguous in va space. */
173         if ((msg->va != st->va) ||
174             ((msg->va+(msg->npages<<PAGE_SHIFT)) > vma->vm_end))
175                 return -EINVAL;
176
177         rc = xen_remap_domain_mfn_range(vma,
178                                         msg->va & PAGE_MASK,
179                                         msg->mfn, msg->npages,
180                                         vma->vm_page_prot,
181                                         st->domain);
182         if (rc < 0)
183                 return rc;
184
185         st->va += msg->npages << PAGE_SHIFT;
186
187         return 0;
188 }
189
190 static long privcmd_ioctl_mmap(void __user *udata)
191 {
192         struct privcmd_mmap mmapcmd;
193         struct mm_struct *mm = current->mm;
194         struct vm_area_struct *vma;
195         int rc;
196         LIST_HEAD(pagelist);
197         struct mmap_mfn_state state;
198
199         if (!xen_initial_domain())
200                 return -EPERM;
201
202         if (copy_from_user(&mmapcmd, udata, sizeof(mmapcmd)))
203                 return -EFAULT;
204
205         rc = gather_array(&pagelist,
206                           mmapcmd.num, sizeof(struct privcmd_mmap_entry),
207                           mmapcmd.entry);
208
209         if (rc || list_empty(&pagelist))
210                 goto out;
211
212         down_write(&mm->mmap_sem);
213
214         {
215                 struct page *page = list_first_entry(&pagelist,
216                                                      struct page, lru);
217                 struct privcmd_mmap_entry *msg = page_address(page);
218
219                 vma = find_vma(mm, msg->va);
220                 rc = -EINVAL;
221
222                 if (!vma || (msg->va != vma->vm_start) ||
223                     !privcmd_enforce_singleshot_mapping(vma))
224                         goto out_up;
225         }
226
227         state.va = vma->vm_start;
228         state.vma = vma;
229         state.domain = mmapcmd.dom;
230
231         rc = traverse_pages(mmapcmd.num, sizeof(struct privcmd_mmap_entry),
232                             &pagelist,
233                             mmap_mfn_range, &state);
234
235
236 out_up:
237         up_write(&mm->mmap_sem);
238
239 out:
240         free_page_list(&pagelist);
241
242         return rc;
243 }
244
245 struct mmap_batch_state {
246         domid_t domain;
247         unsigned long va;
248         struct vm_area_struct *vma;
249         int err;
250
251         xen_pfn_t __user *user;
252 };
253
254 static int mmap_batch_fn(void *data, void *state)
255 {
256         xen_pfn_t *mfnp = data;
257         struct mmap_batch_state *st = state;
258
259         if (xen_remap_domain_mfn_range(st->vma, st->va & PAGE_MASK, *mfnp, 1,
260                                        st->vma->vm_page_prot, st->domain) < 0) {
261                 *mfnp |= 0xf0000000U;
262                 st->err++;
263         }
264         st->va += PAGE_SIZE;
265
266         return 0;
267 }
268
269 static int mmap_return_errors(void *data, void *state)
270 {
271         xen_pfn_t *mfnp = data;
272         struct mmap_batch_state *st = state;
273
274         return put_user(*mfnp, st->user++);
275 }
276
277 static struct vm_operations_struct privcmd_vm_ops;
278
279 static long privcmd_ioctl_mmap_batch(void __user *udata)
280 {
281         int ret;
282         struct privcmd_mmapbatch m;
283         struct mm_struct *mm = current->mm;
284         struct vm_area_struct *vma;
285         unsigned long nr_pages;
286         LIST_HEAD(pagelist);
287         struct mmap_batch_state state;
288
289         if (!xen_initial_domain())
290                 return -EPERM;
291
292         if (copy_from_user(&m, udata, sizeof(m)))
293                 return -EFAULT;
294
295         nr_pages = m.num;
296         if ((m.num <= 0) || (nr_pages > (LONG_MAX >> PAGE_SHIFT)))
297                 return -EINVAL;
298
299         ret = gather_array(&pagelist, m.num, sizeof(xen_pfn_t),
300                            m.arr);
301
302         if (ret || list_empty(&pagelist))
303                 goto out;
304
305         down_write(&mm->mmap_sem);
306
307         vma = find_vma(mm, m.addr);
308         ret = -EINVAL;
309         if (!vma ||
310             vma->vm_ops != &privcmd_vm_ops ||
311             (m.addr != vma->vm_start) ||
312             ((m.addr + (nr_pages << PAGE_SHIFT)) != vma->vm_end) ||
313             !privcmd_enforce_singleshot_mapping(vma)) {
314                 up_write(&mm->mmap_sem);
315                 goto out;
316         }
317
318         state.domain = m.dom;
319         state.vma = vma;
320         state.va = m.addr;
321         state.err = 0;
322
323         ret = traverse_pages(m.num, sizeof(xen_pfn_t),
324                              &pagelist, mmap_batch_fn, &state);
325
326         up_write(&mm->mmap_sem);
327
328         if (state.err > 0) {
329                 state.user = m.arr;
330                 ret = traverse_pages(m.num, sizeof(xen_pfn_t),
331                                &pagelist,
332                                mmap_return_errors, &state);
333         }
334
335 out:
336         free_page_list(&pagelist);
337
338         return ret;
339 }
340
341 static long privcmd_ioctl(struct file *file,
342                           unsigned int cmd, unsigned long data)
343 {
344         int ret = -ENOSYS;
345         void __user *udata = (void __user *) data;
346
347         switch (cmd) {
348         case IOCTL_PRIVCMD_HYPERCALL:
349                 ret = privcmd_ioctl_hypercall(udata);
350                 break;
351
352         case IOCTL_PRIVCMD_MMAP:
353                 ret = privcmd_ioctl_mmap(udata);
354                 break;
355
356         case IOCTL_PRIVCMD_MMAPBATCH:
357                 ret = privcmd_ioctl_mmap_batch(udata);
358                 break;
359
360         default:
361                 ret = -EINVAL;
362                 break;
363         }
364
365         return ret;
366 }
367
368 static int privcmd_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
369 {
370         printk(KERN_DEBUG "privcmd_fault: vma=%p %lx-%lx, pgoff=%lx, uv=%p\n",
371                vma, vma->vm_start, vma->vm_end,
372                vmf->pgoff, vmf->virtual_address);
373
374         return VM_FAULT_SIGBUS;
375 }
376
377 static struct vm_operations_struct privcmd_vm_ops = {
378         .fault = privcmd_fault
379 };
380
381 static int privcmd_mmap(struct file *file, struct vm_area_struct *vma)
382 {
383         /* Unsupported for auto-translate guests. */
384         if (xen_feature(XENFEAT_auto_translated_physmap))
385                 return -ENOSYS;
386
387         /* DONTCOPY is essential for Xen because copy_page_range doesn't know
388          * how to recreate these mappings */
389         vma->vm_flags |= VM_RESERVED | VM_IO | VM_DONTCOPY | VM_PFNMAP;
390         vma->vm_ops = &privcmd_vm_ops;
391         vma->vm_private_data = NULL;
392
393         return 0;
394 }
395
396 static int privcmd_enforce_singleshot_mapping(struct vm_area_struct *vma)
397 {
398         return (xchg(&vma->vm_private_data, (void *)1) == NULL);
399 }
400
401 const struct file_operations xen_privcmd_fops = {
402         .owner = THIS_MODULE,
403         .unlocked_ioctl = privcmd_ioctl,
404         .mmap = privcmd_mmap,
405 };
406 EXPORT_SYMBOL_GPL(xen_privcmd_fops);
407
408 static struct miscdevice privcmd_dev = {
409         .minor = MISC_DYNAMIC_MINOR,
410         .name = "xen/privcmd",
411         .fops = &xen_privcmd_fops,
412 };
413
414 static int __init privcmd_init(void)
415 {
416         int err;
417
418         if (!xen_domain())
419                 return -ENODEV;
420
421         err = misc_register(&privcmd_dev);
422         if (err != 0) {
423                 printk(KERN_ERR "Could not register Xen privcmd device\n");
424                 return err;
425         }
426         return 0;
427 }
428
429 static void __exit privcmd_exit(void)
430 {
431         misc_deregister(&privcmd_dev);
432 }
433
434 module_init(privcmd_init);
435 module_exit(privcmd_exit);