]>
Commit | Line | Data |
---|---|---|
1da177e4 | 1 | /* |
ba180fd4 | 2 | * Copyright (C) 2001 - 2007 Jeff Dike (jdike@{addtoit,linux.intel}.com) |
1da177e4 LT |
3 | * Licensed under the GPL |
4 | */ | |
5 | ||
ba180fd4 | 6 | #include "linux/mm.h" |
858259cf | 7 | #include "asm/unistd.h" |
12919aa6 | 8 | #include "os.h" |
ba180fd4 | 9 | #include "proc_mm.h" |
858259cf BS |
10 | #include "skas.h" |
11 | #include "skas_ptrace.h" | |
ba180fd4 JD |
12 | #include "sysdep/tls.h" |
13 | ||
14 | extern int modify_ldt(int func, void *ptr, unsigned long bytecount); | |
858259cf BS |
15 | |
16 | long write_ldt_entry(struct mm_id * mm_idp, int func, struct user_desc * desc, | |
17 | void **addr, int done) | |
18 | { | |
19 | long res; | |
20 | ||
ba180fd4 JD |
21 | if (proc_mm) { |
22 | /* | |
23 | * This is a special handling for the case, that the mm to | |
858259cf BS |
24 | * modify isn't current->active_mm. |
25 | * If this is called directly by modify_ldt, | |
26 | * (current->active_mm->context.skas.u == mm_idp) | |
77bf4400 | 27 | * will be true. So no call to __switch_mm(mm_idp) is done. |
858259cf BS |
28 | * If this is called in case of init_new_ldt or PTRACE_LDT, |
29 | * mm_idp won't belong to current->active_mm, but child->mm. | |
30 | * So we need to switch child's mm into our userspace, then | |
31 | * later switch back. | |
32 | * | |
07f4e2c6 | 33 | * Note: I'm unsure: should interrupts be disabled here? |
858259cf | 34 | */ |
ba180fd4 | 35 | if (!current->active_mm || current->active_mm == &init_mm || |
6c738ffa | 36 | mm_idp != ¤t->active_mm->context.id) |
77bf4400 | 37 | __switch_mm(mm_idp); |
858259cf BS |
38 | } |
39 | ||
ba180fd4 | 40 | if (ptrace_ldt) { |
858259cf BS |
41 | struct ptrace_ldt ldt_op = (struct ptrace_ldt) { |
42 | .func = func, | |
43 | .ptr = desc, | |
44 | .bytecount = sizeof(*desc)}; | |
45 | u32 cpu; | |
46 | int pid; | |
47 | ||
ba180fd4 | 48 | if (!proc_mm) |
858259cf BS |
49 | pid = mm_idp->u.pid; |
50 | else { | |
51 | cpu = get_cpu(); | |
52 | pid = userspace_pid[cpu]; | |
53 | } | |
54 | ||
07f4e2c6 | 55 | res = os_ptrace_ldt(pid, 0, (unsigned long) &ldt_op); |
858259cf | 56 | |
ba180fd4 | 57 | if (proc_mm) |
858259cf BS |
58 | put_cpu(); |
59 | } | |
60 | else { | |
61 | void *stub_addr; | |
62 | res = syscall_stub_data(mm_idp, (unsigned long *)desc, | |
63 | (sizeof(*desc) + sizeof(long) - 1) & | |
64 | ~(sizeof(long) - 1), | |
65 | addr, &stub_addr); | |
ba180fd4 | 66 | if (!res) { |
858259cf BS |
67 | unsigned long args[] = { func, |
68 | (unsigned long)stub_addr, | |
69 | sizeof(*desc), | |
70 | 0, 0, 0 }; | |
71 | res = run_syscall_stub(mm_idp, __NR_modify_ldt, args, | |
72 | 0, addr, done); | |
73 | } | |
74 | } | |
75 | ||
ba180fd4 JD |
76 | if (proc_mm) { |
77 | /* | |
78 | * This is the second part of special handling, that makes | |
858259cf BS |
79 | * PTRACE_LDT possible to implement. |
80 | */ | |
ba180fd4 | 81 | if (current->active_mm && current->active_mm != &init_mm && |
6c738ffa JD |
82 | mm_idp != ¤t->active_mm->context.id) |
83 | __switch_mm(¤t->active_mm->context.id); | |
858259cf BS |
84 | } |
85 | ||
86 | return res; | |
87 | } | |
88 | ||
89 | static long read_ldt_from_host(void __user * ptr, unsigned long bytecount) | |
90 | { | |
91 | int res, n; | |
92 | struct ptrace_ldt ptrace_ldt = (struct ptrace_ldt) { | |
93 | .func = 0, | |
94 | .bytecount = bytecount, | |
5cbded58 | 95 | .ptr = kmalloc(bytecount, GFP_KERNEL)}; |
858259cf BS |
96 | u32 cpu; |
97 | ||
ba180fd4 | 98 | if (ptrace_ldt.ptr == NULL) |
858259cf BS |
99 | return -ENOMEM; |
100 | ||
ba180fd4 JD |
101 | /* |
102 | * This is called from sys_modify_ldt only, so userspace_pid gives | |
858259cf BS |
103 | * us the right number |
104 | */ | |
105 | ||
106 | cpu = get_cpu(); | |
07f4e2c6 | 107 | res = os_ptrace_ldt(userspace_pid[cpu], 0, (unsigned long) &ptrace_ldt); |
858259cf | 108 | put_cpu(); |
ba180fd4 | 109 | if (res < 0) |
858259cf BS |
110 | goto out; |
111 | ||
112 | n = copy_to_user(ptr, ptrace_ldt.ptr, res); | |
ba180fd4 | 113 | if (n != 0) |
858259cf BS |
114 | res = -EFAULT; |
115 | ||
116 | out: | |
117 | kfree(ptrace_ldt.ptr); | |
118 | ||
119 | return res; | |
120 | } | |
121 | ||
122 | /* | |
123 | * In skas mode, we hold our own ldt data in UML. | |
124 | * Thus, the code implementing sys_modify_ldt_skas | |
125 | * is very similar to (and mostly stolen from) sys_modify_ldt | |
126 | * for arch/i386/kernel/ldt.c | |
127 | * The routines copied and modified in part are: | |
128 | * - read_ldt | |
129 | * - read_default_ldt | |
130 | * - write_ldt | |
131 | * - sys_modify_ldt_skas | |
132 | */ | |
133 | ||
134 | static int read_ldt(void __user * ptr, unsigned long bytecount) | |
135 | { | |
136 | int i, err = 0; | |
137 | unsigned long size; | |
6c738ffa | 138 | uml_ldt_t * ldt = ¤t->mm->context.ldt; |
858259cf | 139 | |
ba180fd4 | 140 | if (!ldt->entry_count) |
858259cf | 141 | goto out; |
ba180fd4 | 142 | if (bytecount > LDT_ENTRY_SIZE*LDT_ENTRIES) |
858259cf BS |
143 | bytecount = LDT_ENTRY_SIZE*LDT_ENTRIES; |
144 | err = bytecount; | |
145 | ||
ba180fd4 | 146 | if (ptrace_ldt) |
858259cf | 147 | return read_ldt_from_host(ptr, bytecount); |
858259cf BS |
148 | |
149 | down(&ldt->semaphore); | |
ba180fd4 | 150 | if (ldt->entry_count <= LDT_DIRECT_ENTRIES) { |
858259cf | 151 | size = LDT_ENTRY_SIZE*LDT_DIRECT_ENTRIES; |
ba180fd4 | 152 | if (size > bytecount) |
858259cf | 153 | size = bytecount; |
ba180fd4 | 154 | if (copy_to_user(ptr, ldt->u.entries, size)) |
858259cf BS |
155 | err = -EFAULT; |
156 | bytecount -= size; | |
157 | ptr += size; | |
158 | } | |
159 | else { | |
ba180fd4 JD |
160 | for (i=0; i<ldt->entry_count/LDT_ENTRIES_PER_PAGE && bytecount; |
161 | i++) { | |
858259cf | 162 | size = PAGE_SIZE; |
ba180fd4 | 163 | if (size > bytecount) |
858259cf | 164 | size = bytecount; |
ba180fd4 | 165 | if (copy_to_user(ptr, ldt->u.pages[i], size)) { |
858259cf BS |
166 | err = -EFAULT; |
167 | break; | |
168 | } | |
169 | bytecount -= size; | |
170 | ptr += size; | |
171 | } | |
172 | } | |
173 | up(&ldt->semaphore); | |
174 | ||
ba180fd4 | 175 | if (bytecount == 0 || err == -EFAULT) |
858259cf BS |
176 | goto out; |
177 | ||
ba180fd4 | 178 | if (clear_user(ptr, bytecount)) |
858259cf BS |
179 | err = -EFAULT; |
180 | ||
181 | out: | |
182 | return err; | |
183 | } | |
184 | ||
185 | static int read_default_ldt(void __user * ptr, unsigned long bytecount) | |
186 | { | |
187 | int err; | |
188 | ||
ba180fd4 | 189 | if (bytecount > 5*LDT_ENTRY_SIZE) |
858259cf BS |
190 | bytecount = 5*LDT_ENTRY_SIZE; |
191 | ||
192 | err = bytecount; | |
ba180fd4 JD |
193 | /* |
194 | * UML doesn't support lcall7 and lcall27. | |
858259cf BS |
195 | * So, we don't really have a default ldt, but emulate |
196 | * an empty ldt of common host default ldt size. | |
197 | */ | |
ba180fd4 | 198 | if (clear_user(ptr, bytecount)) |
858259cf BS |
199 | err = -EFAULT; |
200 | ||
201 | return err; | |
202 | } | |
203 | ||
204 | static int write_ldt(void __user * ptr, unsigned long bytecount, int func) | |
205 | { | |
6c738ffa JD |
206 | uml_ldt_t * ldt = ¤t->mm->context.ldt; |
207 | struct mm_id * mm_idp = ¤t->mm->context.id; | |
858259cf BS |
208 | int i, err; |
209 | struct user_desc ldt_info; | |
210 | struct ldt_entry entry0, *ldt_p; | |
211 | void *addr = NULL; | |
212 | ||
213 | err = -EINVAL; | |
ba180fd4 | 214 | if (bytecount != sizeof(ldt_info)) |
858259cf BS |
215 | goto out; |
216 | err = -EFAULT; | |
ba180fd4 | 217 | if (copy_from_user(&ldt_info, ptr, sizeof(ldt_info))) |
858259cf BS |
218 | goto out; |
219 | ||
220 | err = -EINVAL; | |
ba180fd4 | 221 | if (ldt_info.entry_number >= LDT_ENTRIES) |
858259cf | 222 | goto out; |
ba180fd4 | 223 | if (ldt_info.contents == 3) { |
858259cf BS |
224 | if (func == 1) |
225 | goto out; | |
226 | if (ldt_info.seg_not_present == 0) | |
227 | goto out; | |
228 | } | |
229 | ||
ba180fd4 JD |
230 | if (!ptrace_ldt) |
231 | down(&ldt->semaphore); | |
858259cf BS |
232 | |
233 | err = write_ldt_entry(mm_idp, func, &ldt_info, &addr, 1); | |
ba180fd4 | 234 | if (err) |
858259cf | 235 | goto out_unlock; |
ba180fd4 JD |
236 | else if (ptrace_ldt) { |
237 | /* With PTRACE_LDT available, this is used as a flag only */ | |
238 | ldt->entry_count = 1; | |
239 | goto out; | |
240 | } | |
241 | ||
242 | if (ldt_info.entry_number >= ldt->entry_count && | |
243 | ldt_info.entry_number >= LDT_DIRECT_ENTRIES) { | |
244 | for (i=ldt->entry_count/LDT_ENTRIES_PER_PAGE; | |
245 | i*LDT_ENTRIES_PER_PAGE <= ldt_info.entry_number; | |
246 | i++) { | |
247 | if (i == 0) | |
e23181de JD |
248 | memcpy(&entry0, ldt->u.entries, |
249 | sizeof(entry0)); | |
250 | ldt->u.pages[i] = (struct ldt_entry *) | |
251 | __get_free_page(GFP_KERNEL|__GFP_ZERO); | |
ba180fd4 | 252 | if (!ldt->u.pages[i]) { |
858259cf BS |
253 | err = -ENOMEM; |
254 | /* Undo the change in host */ | |
255 | memset(&ldt_info, 0, sizeof(ldt_info)); | |
256 | write_ldt_entry(mm_idp, 1, &ldt_info, &addr, 1); | |
257 | goto out_unlock; | |
258 | } | |
ba180fd4 | 259 | if (i == 0) { |
e23181de JD |
260 | memcpy(ldt->u.pages[0], &entry0, |
261 | sizeof(entry0)); | |
262 | memcpy(ldt->u.pages[0]+1, ldt->u.entries+1, | |
858259cf BS |
263 | sizeof(entry0)*(LDT_DIRECT_ENTRIES-1)); |
264 | } | |
265 | ldt->entry_count = (i + 1) * LDT_ENTRIES_PER_PAGE; | |
266 | } | |
267 | } | |
ba180fd4 | 268 | if (ldt->entry_count <= ldt_info.entry_number) |
858259cf BS |
269 | ldt->entry_count = ldt_info.entry_number + 1; |
270 | ||
ba180fd4 | 271 | if (ldt->entry_count <= LDT_DIRECT_ENTRIES) |
e23181de | 272 | ldt_p = ldt->u.entries + ldt_info.entry_number; |
858259cf | 273 | else |
e23181de | 274 | ldt_p = ldt->u.pages[ldt_info.entry_number/LDT_ENTRIES_PER_PAGE] + |
858259cf BS |
275 | ldt_info.entry_number%LDT_ENTRIES_PER_PAGE; |
276 | ||
ba180fd4 JD |
277 | if (ldt_info.base_addr == 0 && ldt_info.limit == 0 && |
278 | (func == 1 || LDT_empty(&ldt_info))) { | |
858259cf BS |
279 | ldt_p->a = 0; |
280 | ldt_p->b = 0; | |
281 | } | |
282 | else{ | |
283 | if (func == 1) | |
284 | ldt_info.useable = 0; | |
285 | ldt_p->a = LDT_entry_a(&ldt_info); | |
286 | ldt_p->b = LDT_entry_b(&ldt_info); | |
287 | } | |
288 | err = 0; | |
289 | ||
290 | out_unlock: | |
291 | up(&ldt->semaphore); | |
292 | out: | |
293 | return err; | |
294 | } | |
295 | ||
296 | static long do_modify_ldt_skas(int func, void __user *ptr, | |
297 | unsigned long bytecount) | |
298 | { | |
299 | int ret = -ENOSYS; | |
300 | ||
301 | switch (func) { | |
302 | case 0: | |
303 | ret = read_ldt(ptr, bytecount); | |
304 | break; | |
305 | case 1: | |
306 | case 0x11: | |
307 | ret = write_ldt(ptr, bytecount, func); | |
308 | break; | |
309 | case 2: | |
310 | ret = read_default_ldt(ptr, bytecount); | |
311 | break; | |
312 | } | |
313 | return ret; | |
314 | } | |
315 | ||
af727902 JD |
316 | static DEFINE_SPINLOCK(host_ldt_lock); |
317 | static short dummy_list[9] = {0, -1}; | |
318 | static short * host_ldt_entries = NULL; | |
858259cf | 319 | |
af727902 | 320 | static void ldt_get_host_info(void) |
858259cf BS |
321 | { |
322 | long ret; | |
622e6969 JD |
323 | struct ldt_entry * ldt; |
324 | short *tmp; | |
858259cf BS |
325 | int i, size, k, order; |
326 | ||
af727902 JD |
327 | spin_lock(&host_ldt_lock); |
328 | ||
ba180fd4 | 329 | if (host_ldt_entries != NULL) { |
af727902 JD |
330 | spin_unlock(&host_ldt_lock); |
331 | return; | |
332 | } | |
858259cf BS |
333 | host_ldt_entries = dummy_list+1; |
334 | ||
af727902 JD |
335 | spin_unlock(&host_ldt_lock); |
336 | ||
ba180fd4 JD |
337 | for (i = LDT_PAGES_MAX-1, order=0; i; i>>=1, order++) |
338 | ; | |
858259cf BS |
339 | |
340 | ldt = (struct ldt_entry *) | |
341 | __get_free_pages(GFP_KERNEL|__GFP_ZERO, order); | |
ba180fd4 JD |
342 | if (ldt == NULL) { |
343 | printk(KERN_ERR "ldt_get_host_info: couldn't allocate buffer " | |
344 | "for host ldt\n"); | |
858259cf BS |
345 | return; |
346 | } | |
347 | ||
348 | ret = modify_ldt(0, ldt, (1<<order)*PAGE_SIZE); | |
ba180fd4 JD |
349 | if (ret < 0) { |
350 | printk(KERN_ERR "ldt_get_host_info: couldn't read host ldt\n"); | |
858259cf BS |
351 | goto out_free; |
352 | } | |
ba180fd4 | 353 | if (ret == 0) { |
858259cf BS |
354 | /* default_ldt is active, simply write an empty entry 0 */ |
355 | host_ldt_entries = dummy_list; | |
356 | goto out_free; | |
357 | } | |
358 | ||
ba180fd4 JD |
359 | for (i=0, size=0; i<ret/LDT_ENTRY_SIZE; i++) { |
360 | if (ldt[i].a != 0 || ldt[i].b != 0) | |
858259cf BS |
361 | size++; |
362 | } | |
363 | ||
ba180fd4 | 364 | if (size < ARRAY_SIZE(dummy_list)) |
858259cf | 365 | host_ldt_entries = dummy_list; |
858259cf BS |
366 | else { |
367 | size = (size + 1) * sizeof(dummy_list[0]); | |
af727902 | 368 | tmp = kmalloc(size, GFP_KERNEL); |
ba180fd4 JD |
369 | if (tmp == NULL) { |
370 | printk(KERN_ERR "ldt_get_host_info: couldn't allocate " | |
371 | "host ldt list\n"); | |
858259cf BS |
372 | goto out_free; |
373 | } | |
af727902 | 374 | host_ldt_entries = tmp; |
858259cf BS |
375 | } |
376 | ||
ba180fd4 JD |
377 | for (i=0, k=0; i<ret/LDT_ENTRY_SIZE; i++) { |
378 | if (ldt[i].a != 0 || ldt[i].b != 0) | |
858259cf | 379 | host_ldt_entries[k++] = i; |
858259cf BS |
380 | } |
381 | host_ldt_entries[k] = -1; | |
382 | ||
383 | out_free: | |
384 | free_pages((unsigned long)ldt, order); | |
385 | } | |
386 | ||
6c738ffa | 387 | long init_new_ldt(struct mm_context *new_mm, struct mm_context *from_mm) |
858259cf BS |
388 | { |
389 | struct user_desc desc; | |
390 | short * num_p; | |
391 | int i; | |
392 | long page, err=0; | |
393 | void *addr = NULL; | |
12919aa6 | 394 | struct proc_mm_op copy; |
858259cf | 395 | |
858259cf | 396 | |
ba180fd4 | 397 | if (!ptrace_ldt) |
858259cf BS |
398 | init_MUTEX(&new_mm->ldt.semaphore); |
399 | ||
ba180fd4 | 400 | if (!from_mm) { |
12919aa6 | 401 | memset(&desc, 0, sizeof(desc)); |
858259cf BS |
402 | /* |
403 | * We have to initialize a clean ldt. | |
404 | */ | |
ba180fd4 | 405 | if (proc_mm) { |
858259cf BS |
406 | /* |
407 | * If the new mm was created using proc_mm, host's | |
408 | * default-ldt currently is assigned, which normally | |
409 | * contains the call-gates for lcall7 and lcall27. | |
410 | * To remove these gates, we simply write an empty | |
411 | * entry as number 0 to the host. | |
412 | */ | |
ba180fd4 | 413 | err = write_ldt_entry(&new_mm->id, 1, &desc, &addr, 1); |
858259cf BS |
414 | } |
415 | else{ | |
416 | /* | |
417 | * Now we try to retrieve info about the ldt, we | |
418 | * inherited from the host. All ldt-entries found | |
419 | * will be reset in the following loop | |
420 | */ | |
af727902 | 421 | ldt_get_host_info(); |
ba180fd4 | 422 | for (num_p=host_ldt_entries; *num_p != -1; num_p++) { |
858259cf BS |
423 | desc.entry_number = *num_p; |
424 | err = write_ldt_entry(&new_mm->id, 1, &desc, | |
425 | &addr, *(num_p + 1) == -1); | |
ba180fd4 | 426 | if (err) |
858259cf BS |
427 | break; |
428 | } | |
429 | } | |
430 | new_mm->ldt.entry_count = 0; | |
12919aa6 BS |
431 | |
432 | goto out; | |
858259cf | 433 | } |
12919aa6 | 434 | |
ba180fd4 JD |
435 | if (proc_mm) { |
436 | /* | |
437 | * We have a valid from_mm, so we now have to copy the LDT of | |
12919aa6 BS |
438 | * from_mm to new_mm, because using proc_mm an new mm with |
439 | * an empty/default LDT was created in new_mm() | |
440 | */ | |
441 | copy = ((struct proc_mm_op) { .op = MM_COPY_SEGMENTS, | |
442 | .u = | |
443 | { .copy_segments = | |
444 | from_mm->id.u.mm_fd } } ); | |
a6ea4cce | 445 | i = os_write_file(new_mm->id.u.mm_fd, ©, sizeof(copy)); |
ba180fd4 JD |
446 | if (i != sizeof(copy)) |
447 | printk(KERN_ERR "new_mm : /proc/mm copy_segments " | |
448 | "failed, err = %d\n", -i); | |
12919aa6 BS |
449 | } |
450 | ||
ba180fd4 JD |
451 | if (!ptrace_ldt) { |
452 | /* | |
453 | * Our local LDT is used to supply the data for | |
858259cf BS |
454 | * modify_ldt(READLDT), if PTRACE_LDT isn't available, |
455 | * i.e., we have to use the stub for modify_ldt, which | |
456 | * can't handle the big read buffer of up to 64kB. | |
457 | */ | |
458 | down(&from_mm->ldt.semaphore); | |
ba180fd4 | 459 | if (from_mm->ldt.entry_count <= LDT_DIRECT_ENTRIES) |
e23181de JD |
460 | memcpy(new_mm->ldt.u.entries, from_mm->ldt.u.entries, |
461 | sizeof(new_mm->ldt.u.entries)); | |
ba180fd4 | 462 | else { |
858259cf | 463 | i = from_mm->ldt.entry_count / LDT_ENTRIES_PER_PAGE; |
ba180fd4 | 464 | while (i-->0) { |
858259cf | 465 | page = __get_free_page(GFP_KERNEL|__GFP_ZERO); |
ba180fd4 | 466 | if (!page) { |
858259cf BS |
467 | err = -ENOMEM; |
468 | break; | |
469 | } | |
e23181de JD |
470 | new_mm->ldt.u.pages[i] = |
471 | (struct ldt_entry *) page; | |
472 | memcpy(new_mm->ldt.u.pages[i], | |
473 | from_mm->ldt.u.pages[i], PAGE_SIZE); | |
858259cf BS |
474 | } |
475 | } | |
476 | new_mm->ldt.entry_count = from_mm->ldt.entry_count; | |
477 | up(&from_mm->ldt.semaphore); | |
478 | } | |
479 | ||
12919aa6 | 480 | out: |
858259cf BS |
481 | return err; |
482 | } | |
483 | ||
484 | ||
6c738ffa | 485 | void free_ldt(struct mm_context *mm) |
858259cf BS |
486 | { |
487 | int i; | |
488 | ||
ba180fd4 | 489 | if (!ptrace_ldt && mm->ldt.entry_count > LDT_DIRECT_ENTRIES) { |
858259cf | 490 | i = mm->ldt.entry_count / LDT_ENTRIES_PER_PAGE; |
ba180fd4 JD |
491 | while (i-- > 0) |
492 | free_page((long) mm->ldt.u.pages[i]); | |
858259cf BS |
493 | } |
494 | mm->ldt.entry_count = 0; | |
495 | } | |
858259cf BS |
496 | |
497 | int sys_modify_ldt(int func, void __user *ptr, unsigned long bytecount) | |
498 | { | |
6aa802ce | 499 | return do_modify_ldt_skas(func, ptr, bytecount); |
858259cf | 500 | } |