]>
Commit | Line | Data |
---|---|---|
1c5de193 JF |
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/sched.h> | |
11 | #include <linux/slab.h> | |
12 | #include <linux/string.h> | |
13 | #include <linux/errno.h> | |
14 | #include <linux/mm.h> | |
15 | #include <linux/mman.h> | |
16 | #include <linux/uaccess.h> | |
17 | #include <linux/swap.h> | |
18 | #include <linux/smp_lock.h> | |
19 | #include <linux/highmem.h> | |
20 | #include <linux/pagemap.h> | |
21 | #include <linux/seq_file.h> | |
22 | ||
23 | #include <asm/pgalloc.h> | |
24 | #include <asm/pgtable.h> | |
25 | #include <asm/tlb.h> | |
26 | #include <asm/xen/hypervisor.h> | |
27 | #include <asm/xen/hypercall.h> | |
28 | ||
29 | #include <xen/xen.h> | |
30 | #include <xen/privcmd.h> | |
31 | #include <xen/interface/xen.h> | |
32 | #include <xen/features.h> | |
33 | #include <xen/page.h> | |
34 | ||
f020e290 IC |
35 | #define REMAP_BATCH_SIZE 16 |
36 | ||
1c5de193 JF |
37 | #ifndef HAVE_ARCH_PRIVCMD_MMAP |
38 | static int privcmd_enforce_singleshot_mapping(struct vm_area_struct *vma); | |
39 | #endif | |
40 | ||
41 | struct remap_data { | |
42 | unsigned long mfn; | |
1c5de193 | 43 | pgprot_t prot; |
f020e290 | 44 | struct mmu_update *mmu_update; |
1c5de193 JF |
45 | }; |
46 | ||
47 | static int remap_area_mfn_pte_fn(pte_t *ptep, pgtable_t token, | |
48 | unsigned long addr, void *data) | |
49 | { | |
50 | struct remap_data *rmd = data; | |
51 | pte_t pte = pte_mkspecial(pfn_pte(rmd->mfn++, rmd->prot)); | |
52 | ||
f020e290 IC |
53 | rmd->mmu_update->ptr = arbitrary_virt_to_machine(ptep).maddr; |
54 | rmd->mmu_update->val = pte_val_ma(pte); | |
55 | rmd->mmu_update++; | |
1c5de193 JF |
56 | |
57 | return 0; | |
58 | } | |
59 | ||
f020e290 IC |
60 | static int remap_domain_mfn_range(struct vm_area_struct *vma, |
61 | unsigned long addr, | |
62 | unsigned long mfn, int nr, | |
63 | pgprot_t prot, unsigned domid) | |
1c5de193 JF |
64 | { |
65 | struct remap_data rmd; | |
f020e290 IC |
66 | struct mmu_update mmu_update[REMAP_BATCH_SIZE]; |
67 | int batch; | |
68 | unsigned long range; | |
69 | int err = 0; | |
1c5de193 JF |
70 | |
71 | prot = __pgprot(pgprot_val(prot) | _PAGE_IOMAP); | |
72 | ||
73 | vma->vm_flags |= VM_IO | VM_RESERVED | VM_PFNMAP; | |
74 | ||
75 | rmd.mfn = mfn; | |
76 | rmd.prot = prot; | |
1c5de193 | 77 | |
f020e290 IC |
78 | while (nr) { |
79 | batch = min(REMAP_BATCH_SIZE, nr); | |
80 | range = (unsigned long)batch << PAGE_SHIFT; | |
81 | ||
82 | rmd.mmu_update = mmu_update; | |
83 | err = apply_to_page_range(vma->vm_mm, addr, range, | |
84 | remap_area_mfn_pte_fn, &rmd); | |
85 | if (err) | |
86 | goto out; | |
87 | ||
88 | err = -EFAULT; | |
89 | if (HYPERVISOR_mmu_update(mmu_update, batch, NULL, domid) < 0) | |
90 | goto out; | |
91 | ||
92 | nr -= batch; | |
93 | addr += range; | |
94 | } | |
95 | ||
96 | err = 0; | |
97 | out: | |
98 | ||
99 | flush_tlb_all(); | |
1c5de193 JF |
100 | |
101 | return err; | |
102 | } | |
103 | ||
104 | static long privcmd_ioctl_hypercall(void __user *udata) | |
105 | { | |
106 | struct privcmd_hypercall hypercall; | |
107 | long ret; | |
108 | ||
109 | if (copy_from_user(&hypercall, udata, sizeof(hypercall))) | |
110 | return -EFAULT; | |
111 | ||
112 | ret = privcmd_call(hypercall.op, | |
113 | hypercall.arg[0], hypercall.arg[1], | |
114 | hypercall.arg[2], hypercall.arg[3], | |
115 | hypercall.arg[4]); | |
116 | ||
117 | return ret; | |
118 | } | |
119 | ||
120 | static void free_page_list(struct list_head *pages) | |
121 | { | |
122 | struct page *p, *n; | |
123 | ||
124 | list_for_each_entry_safe(p, n, pages, lru) | |
125 | __free_page(p); | |
126 | ||
127 | INIT_LIST_HEAD(pages); | |
128 | } | |
129 | ||
130 | /* | |
131 | * Given an array of items in userspace, return a list of pages | |
132 | * containing the data. If copying fails, either because of memory | |
133 | * allocation failure or a problem reading user memory, return an | |
134 | * error code; its up to the caller to dispose of any partial list. | |
135 | */ | |
136 | static int gather_array(struct list_head *pagelist, | |
137 | unsigned nelem, size_t size, | |
138 | void __user *data) | |
139 | { | |
140 | unsigned pageidx; | |
141 | void *pagedata; | |
142 | int ret; | |
143 | ||
144 | if (size > PAGE_SIZE) | |
145 | return 0; | |
146 | ||
147 | pageidx = PAGE_SIZE; | |
148 | pagedata = NULL; /* quiet, gcc */ | |
149 | while (nelem--) { | |
150 | if (pageidx > PAGE_SIZE-size) { | |
151 | struct page *page = alloc_page(GFP_KERNEL); | |
152 | ||
153 | ret = -ENOMEM; | |
154 | if (page == NULL) | |
155 | goto fail; | |
156 | ||
157 | pagedata = page_address(page); | |
158 | ||
159 | list_add_tail(&page->lru, pagelist); | |
160 | pageidx = 0; | |
161 | } | |
162 | ||
163 | ret = -EFAULT; | |
164 | if (copy_from_user(pagedata + pageidx, data, size)) | |
165 | goto fail; | |
166 | ||
167 | data += size; | |
168 | pageidx += size; | |
169 | } | |
170 | ||
171 | ret = 0; | |
172 | ||
173 | fail: | |
174 | return ret; | |
175 | } | |
176 | ||
177 | /* | |
178 | * Call function "fn" on each element of the array fragmented | |
179 | * over a list of pages. | |
180 | */ | |
181 | static int traverse_pages(unsigned nelem, size_t size, | |
182 | struct list_head *pos, | |
183 | int (*fn)(void *data, void *state), | |
184 | void *state) | |
185 | { | |
186 | void *pagedata; | |
187 | unsigned pageidx; | |
f020e290 | 188 | int ret = 0; |
1c5de193 JF |
189 | |
190 | BUG_ON(size > PAGE_SIZE); | |
191 | ||
192 | pageidx = PAGE_SIZE; | |
193 | pagedata = NULL; /* hush, gcc */ | |
194 | ||
195 | while (nelem--) { | |
196 | if (pageidx > PAGE_SIZE-size) { | |
197 | struct page *page; | |
198 | pos = pos->next; | |
199 | page = list_entry(pos, struct page, lru); | |
200 | pagedata = page_address(page); | |
201 | pageidx = 0; | |
202 | } | |
203 | ||
204 | ret = (*fn)(pagedata + pageidx, state); | |
205 | if (ret) | |
206 | break; | |
207 | pageidx += size; | |
208 | } | |
209 | ||
210 | return ret; | |
211 | } | |
212 | ||
213 | struct mmap_mfn_state { | |
214 | unsigned long va; | |
215 | struct vm_area_struct *vma; | |
216 | domid_t domain; | |
217 | }; | |
218 | ||
219 | static int mmap_mfn_range(void *data, void *state) | |
220 | { | |
221 | struct privcmd_mmap_entry *msg = data; | |
222 | struct mmap_mfn_state *st = state; | |
223 | struct vm_area_struct *vma = st->vma; | |
224 | int rc; | |
225 | ||
226 | /* Do not allow range to wrap the address space. */ | |
227 | if ((msg->npages > (LONG_MAX >> PAGE_SHIFT)) || | |
228 | ((unsigned long)(msg->npages << PAGE_SHIFT) >= -st->va)) | |
229 | return -EINVAL; | |
230 | ||
231 | /* Range chunks must be contiguous in va space. */ | |
232 | if ((msg->va != st->va) || | |
233 | ((msg->va+(msg->npages<<PAGE_SHIFT)) > vma->vm_end)) | |
234 | return -EINVAL; | |
235 | ||
236 | rc = remap_domain_mfn_range(vma, | |
237 | msg->va & PAGE_MASK, | |
f020e290 | 238 | msg->mfn, msg->npages, |
1c5de193 JF |
239 | vma->vm_page_prot, |
240 | st->domain); | |
241 | if (rc < 0) | |
242 | return rc; | |
243 | ||
244 | st->va += msg->npages << PAGE_SHIFT; | |
245 | ||
246 | return 0; | |
247 | } | |
248 | ||
249 | static long privcmd_ioctl_mmap(void __user *udata) | |
250 | { | |
251 | struct privcmd_mmap mmapcmd; | |
252 | struct mm_struct *mm = current->mm; | |
253 | struct vm_area_struct *vma; | |
254 | int rc; | |
255 | LIST_HEAD(pagelist); | |
256 | struct mmap_mfn_state state; | |
257 | ||
258 | if (!xen_initial_domain()) | |
259 | return -EPERM; | |
260 | ||
261 | if (copy_from_user(&mmapcmd, udata, sizeof(mmapcmd))) | |
262 | return -EFAULT; | |
263 | ||
264 | rc = gather_array(&pagelist, | |
265 | mmapcmd.num, sizeof(struct privcmd_mmap_entry), | |
266 | mmapcmd.entry); | |
267 | ||
268 | if (rc || list_empty(&pagelist)) | |
269 | goto out; | |
270 | ||
271 | down_write(&mm->mmap_sem); | |
272 | ||
273 | { | |
274 | struct page *page = list_first_entry(&pagelist, | |
275 | struct page, lru); | |
276 | struct privcmd_mmap_entry *msg = page_address(page); | |
277 | ||
278 | vma = find_vma(mm, msg->va); | |
279 | rc = -EINVAL; | |
280 | ||
281 | if (!vma || (msg->va != vma->vm_start) || | |
282 | !privcmd_enforce_singleshot_mapping(vma)) | |
283 | goto out_up; | |
284 | } | |
285 | ||
286 | state.va = vma->vm_start; | |
287 | state.vma = vma; | |
288 | state.domain = mmapcmd.dom; | |
289 | ||
290 | rc = traverse_pages(mmapcmd.num, sizeof(struct privcmd_mmap_entry), | |
291 | &pagelist, | |
292 | mmap_mfn_range, &state); | |
293 | ||
294 | ||
295 | out_up: | |
296 | up_write(&mm->mmap_sem); | |
297 | ||
298 | out: | |
299 | free_page_list(&pagelist); | |
300 | ||
301 | return rc; | |
302 | } | |
303 | ||
304 | struct mmap_batch_state { | |
305 | domid_t domain; | |
306 | unsigned long va; | |
307 | struct vm_area_struct *vma; | |
308 | int err; | |
309 | ||
310 | xen_pfn_t __user *user; | |
311 | }; | |
312 | ||
313 | static int mmap_batch_fn(void *data, void *state) | |
314 | { | |
315 | xen_pfn_t *mfnp = data; | |
316 | struct mmap_batch_state *st = state; | |
317 | ||
318 | if (remap_domain_mfn_range(st->vma, st->va & PAGE_MASK, | |
f020e290 | 319 | *mfnp, 1, |
1c5de193 JF |
320 | st->vma->vm_page_prot, st->domain) < 0) { |
321 | *mfnp |= 0xf0000000U; | |
322 | st->err++; | |
323 | } | |
324 | st->va += PAGE_SIZE; | |
325 | ||
326 | return 0; | |
327 | } | |
328 | ||
329 | static int mmap_return_errors(void *data, void *state) | |
330 | { | |
331 | xen_pfn_t *mfnp = data; | |
332 | struct mmap_batch_state *st = state; | |
333 | ||
334 | put_user(*mfnp, st->user++); | |
335 | ||
336 | return 0; | |
337 | } | |
338 | ||
f31fdf51 JF |
339 | static struct vm_operations_struct privcmd_vm_ops; |
340 | ||
1c5de193 JF |
341 | static long privcmd_ioctl_mmap_batch(void __user *udata) |
342 | { | |
343 | int ret; | |
344 | struct privcmd_mmapbatch m; | |
345 | struct mm_struct *mm = current->mm; | |
346 | struct vm_area_struct *vma; | |
347 | unsigned long nr_pages; | |
348 | LIST_HEAD(pagelist); | |
349 | struct mmap_batch_state state; | |
350 | ||
351 | if (!xen_initial_domain()) | |
352 | return -EPERM; | |
353 | ||
354 | if (copy_from_user(&m, udata, sizeof(m))) | |
355 | return -EFAULT; | |
356 | ||
357 | nr_pages = m.num; | |
358 | if ((m.num <= 0) || (nr_pages > (LONG_MAX >> PAGE_SHIFT))) | |
359 | return -EINVAL; | |
360 | ||
361 | ret = gather_array(&pagelist, m.num, sizeof(xen_pfn_t), | |
362 | m.arr); | |
363 | ||
364 | if (ret || list_empty(&pagelist)) | |
365 | goto out; | |
366 | ||
367 | down_write(&mm->mmap_sem); | |
368 | ||
369 | vma = find_vma(mm, m.addr); | |
370 | ret = -EINVAL; | |
371 | if (!vma || | |
f31fdf51 | 372 | vma->vm_ops != &privcmd_vm_ops || |
1c5de193 JF |
373 | (m.addr != vma->vm_start) || |
374 | ((m.addr + (nr_pages << PAGE_SHIFT)) != vma->vm_end) || | |
375 | !privcmd_enforce_singleshot_mapping(vma)) { | |
376 | up_write(&mm->mmap_sem); | |
377 | goto out; | |
378 | } | |
379 | ||
380 | state.domain = m.dom; | |
381 | state.vma = vma; | |
382 | state.va = m.addr; | |
383 | state.err = 0; | |
384 | ||
385 | ret = traverse_pages(m.num, sizeof(xen_pfn_t), | |
386 | &pagelist, mmap_batch_fn, &state); | |
387 | ||
388 | up_write(&mm->mmap_sem); | |
389 | ||
390 | if (state.err > 0) { | |
f020e290 | 391 | ret = 0; |
1c5de193 | 392 | |
f020e290 | 393 | state.user = m.arr; |
1c5de193 JF |
394 | traverse_pages(m.num, sizeof(xen_pfn_t), |
395 | &pagelist, | |
396 | mmap_return_errors, &state); | |
397 | } | |
398 | ||
399 | out: | |
400 | free_page_list(&pagelist); | |
401 | ||
402 | return ret; | |
403 | } | |
404 | ||
405 | static long privcmd_ioctl(struct file *file, | |
406 | unsigned int cmd, unsigned long data) | |
407 | { | |
408 | int ret = -ENOSYS; | |
409 | void __user *udata = (void __user *) data; | |
410 | ||
411 | switch (cmd) { | |
412 | case IOCTL_PRIVCMD_HYPERCALL: | |
413 | ret = privcmd_ioctl_hypercall(udata); | |
414 | break; | |
415 | ||
416 | case IOCTL_PRIVCMD_MMAP: | |
417 | ret = privcmd_ioctl_mmap(udata); | |
418 | break; | |
419 | ||
420 | case IOCTL_PRIVCMD_MMAPBATCH: | |
421 | ret = privcmd_ioctl_mmap_batch(udata); | |
422 | break; | |
423 | ||
424 | default: | |
425 | ret = -EINVAL; | |
426 | break; | |
427 | } | |
428 | ||
429 | return ret; | |
430 | } | |
431 | ||
432 | #ifndef HAVE_ARCH_PRIVCMD_MMAP | |
433 | static int privcmd_fault(struct vm_area_struct *vma, struct vm_fault *vmf) | |
434 | { | |
441c7416 JF |
435 | printk(KERN_DEBUG "privcmd_fault: vma=%p %lx-%lx, pgoff=%lx, uv=%p\n", |
436 | vma, vma->vm_start, vma->vm_end, | |
437 | vmf->pgoff, vmf->virtual_address); | |
438 | ||
1c5de193 JF |
439 | return VM_FAULT_SIGBUS; |
440 | } | |
441 | ||
442 | static struct vm_operations_struct privcmd_vm_ops = { | |
443 | .fault = privcmd_fault | |
444 | }; | |
445 | ||
446 | static int privcmd_mmap(struct file *file, struct vm_area_struct *vma) | |
447 | { | |
448 | /* Unsupported for auto-translate guests. */ | |
449 | if (xen_feature(XENFEAT_auto_translated_physmap)) | |
450 | return -ENOSYS; | |
451 | ||
452 | /* DONTCOPY is essential for Xen as copy_page_range is broken. */ | |
453 | vma->vm_flags |= VM_RESERVED | VM_IO | VM_DONTCOPY; | |
454 | vma->vm_ops = &privcmd_vm_ops; | |
455 | vma->vm_private_data = NULL; | |
456 | ||
457 | return 0; | |
458 | } | |
459 | ||
460 | static int privcmd_enforce_singleshot_mapping(struct vm_area_struct *vma) | |
461 | { | |
462 | return (xchg(&vma->vm_private_data, (void *)1) == NULL); | |
463 | } | |
464 | #endif | |
465 | ||
466 | const struct file_operations privcmd_file_ops = { | |
467 | .unlocked_ioctl = privcmd_ioctl, | |
468 | .mmap = privcmd_mmap, | |
469 | }; |