Merge commit 'v2.6.27-rc1' into for-linus
[linux-2.6] / arch / um / sys-i386 / ldt.c
1 /*
2  * Copyright (C) 2001 - 2007 Jeff Dike (jdike@{addtoit,linux.intel}.com)
3  * Licensed under the GPL
4  */
5
6 #include <linux/mm.h>
7 #include <linux/sched.h>
8 #include <asm/unistd.h>
9 #include "os.h"
10 #include "proc_mm.h"
11 #include "skas.h"
12 #include "skas_ptrace.h"
13 #include "sysdep/tls.h"
14
15 extern int modify_ldt(int func, void *ptr, unsigned long bytecount);
16
17 static long write_ldt_entry(struct mm_id *mm_idp, int func,
18                      struct user_desc *desc, void **addr, int done)
19 {
20         long res;
21
22         if (proc_mm) {
23                 /*
24                  * This is a special handling for the case, that the mm to
25                  * modify isn't current->active_mm.
26                  * If this is called directly by modify_ldt,
27                  *     (current->active_mm->context.skas.u == mm_idp)
28                  * will be true. So no call to __switch_mm(mm_idp) is done.
29                  * If this is called in case of init_new_ldt or PTRACE_LDT,
30                  * mm_idp won't belong to current->active_mm, but child->mm.
31                  * So we need to switch child's mm into our userspace, then
32                  * later switch back.
33                  *
34                  * Note: I'm unsure: should interrupts be disabled here?
35                  */
36                 if (!current->active_mm || current->active_mm == &init_mm ||
37                     mm_idp != &current->active_mm->context.id)
38                         __switch_mm(mm_idp);
39         }
40
41         if (ptrace_ldt) {
42                 struct ptrace_ldt ldt_op = (struct ptrace_ldt) {
43                         .func = func,
44                         .ptr = desc,
45                         .bytecount = sizeof(*desc)};
46                 u32 cpu;
47                 int pid;
48
49                 if (!proc_mm)
50                         pid = mm_idp->u.pid;
51                 else {
52                         cpu = get_cpu();
53                         pid = userspace_pid[cpu];
54                 }
55
56                 res = os_ptrace_ldt(pid, 0, (unsigned long) &ldt_op);
57
58                 if (proc_mm)
59                         put_cpu();
60         }
61         else {
62                 void *stub_addr;
63                 res = syscall_stub_data(mm_idp, (unsigned long *)desc,
64                                         (sizeof(*desc) + sizeof(long) - 1) &
65                                             ~(sizeof(long) - 1),
66                                         addr, &stub_addr);
67                 if (!res) {
68                         unsigned long args[] = { func,
69                                                  (unsigned long)stub_addr,
70                                                  sizeof(*desc),
71                                                  0, 0, 0 };
72                         res = run_syscall_stub(mm_idp, __NR_modify_ldt, args,
73                                                0, addr, done);
74                 }
75         }
76
77         if (proc_mm) {
78                 /*
79                  * This is the second part of special handling, that makes
80                  * PTRACE_LDT possible to implement.
81                  */
82                 if (current->active_mm && current->active_mm != &init_mm &&
83                     mm_idp != &current->active_mm->context.id)
84                         __switch_mm(&current->active_mm->context.id);
85         }
86
87         return res;
88 }
89
90 static long read_ldt_from_host(void __user * ptr, unsigned long bytecount)
91 {
92         int res, n;
93         struct ptrace_ldt ptrace_ldt = (struct ptrace_ldt) {
94                         .func = 0,
95                         .bytecount = bytecount,
96                         .ptr = kmalloc(bytecount, GFP_KERNEL)};
97         u32 cpu;
98
99         if (ptrace_ldt.ptr == NULL)
100                 return -ENOMEM;
101
102         /*
103          * This is called from sys_modify_ldt only, so userspace_pid gives
104          * us the right number
105          */
106
107         cpu = get_cpu();
108         res = os_ptrace_ldt(userspace_pid[cpu], 0, (unsigned long) &ptrace_ldt);
109         put_cpu();
110         if (res < 0)
111                 goto out;
112
113         n = copy_to_user(ptr, ptrace_ldt.ptr, res);
114         if (n != 0)
115                 res = -EFAULT;
116
117   out:
118         kfree(ptrace_ldt.ptr);
119
120         return res;
121 }
122
123 /*
124  * In skas mode, we hold our own ldt data in UML.
125  * Thus, the code implementing sys_modify_ldt_skas
126  * is very similar to (and mostly stolen from) sys_modify_ldt
127  * for arch/i386/kernel/ldt.c
128  * The routines copied and modified in part are:
129  * - read_ldt
130  * - read_default_ldt
131  * - write_ldt
132  * - sys_modify_ldt_skas
133  */
134
135 static int read_ldt(void __user * ptr, unsigned long bytecount)
136 {
137         int i, err = 0;
138         unsigned long size;
139         uml_ldt_t * ldt = &current->mm->context.ldt;
140
141         if (!ldt->entry_count)
142                 goto out;
143         if (bytecount > LDT_ENTRY_SIZE*LDT_ENTRIES)
144                 bytecount = LDT_ENTRY_SIZE*LDT_ENTRIES;
145         err = bytecount;
146
147         if (ptrace_ldt)
148                 return read_ldt_from_host(ptr, bytecount);
149
150         mutex_lock(&ldt->lock);
151         if (ldt->entry_count <= LDT_DIRECT_ENTRIES) {
152                 size = LDT_ENTRY_SIZE*LDT_DIRECT_ENTRIES;
153                 if (size > bytecount)
154                         size = bytecount;
155                 if (copy_to_user(ptr, ldt->u.entries, size))
156                         err = -EFAULT;
157                 bytecount -= size;
158                 ptr += size;
159         }
160         else {
161                 for (i=0; i<ldt->entry_count/LDT_ENTRIES_PER_PAGE && bytecount;
162                      i++) {
163                         size = PAGE_SIZE;
164                         if (size > bytecount)
165                                 size = bytecount;
166                         if (copy_to_user(ptr, ldt->u.pages[i], size)) {
167                                 err = -EFAULT;
168                                 break;
169                         }
170                         bytecount -= size;
171                         ptr += size;
172                 }
173         }
174         mutex_unlock(&ldt->lock);
175
176         if (bytecount == 0 || err == -EFAULT)
177                 goto out;
178
179         if (clear_user(ptr, bytecount))
180                 err = -EFAULT;
181
182 out:
183         return err;
184 }
185
186 static int read_default_ldt(void __user * ptr, unsigned long bytecount)
187 {
188         int err;
189
190         if (bytecount > 5*LDT_ENTRY_SIZE)
191                 bytecount = 5*LDT_ENTRY_SIZE;
192
193         err = bytecount;
194         /*
195          * UML doesn't support lcall7 and lcall27.
196          * So, we don't really have a default ldt, but emulate
197          * an empty ldt of common host default ldt size.
198          */
199         if (clear_user(ptr, bytecount))
200                 err = -EFAULT;
201
202         return err;
203 }
204
205 static int write_ldt(void __user * ptr, unsigned long bytecount, int func)
206 {
207         uml_ldt_t * ldt = &current->mm->context.ldt;
208         struct mm_id * mm_idp = &current->mm->context.id;
209         int i, err;
210         struct user_desc ldt_info;
211         struct ldt_entry entry0, *ldt_p;
212         void *addr = NULL;
213
214         err = -EINVAL;
215         if (bytecount != sizeof(ldt_info))
216                 goto out;
217         err = -EFAULT;
218         if (copy_from_user(&ldt_info, ptr, sizeof(ldt_info)))
219                 goto out;
220
221         err = -EINVAL;
222         if (ldt_info.entry_number >= LDT_ENTRIES)
223                 goto out;
224         if (ldt_info.contents == 3) {
225                 if (func == 1)
226                         goto out;
227                 if (ldt_info.seg_not_present == 0)
228                         goto out;
229         }
230
231         if (!ptrace_ldt)
232                 mutex_lock(&ldt->lock);
233
234         err = write_ldt_entry(mm_idp, func, &ldt_info, &addr, 1);
235         if (err)
236                 goto out_unlock;
237         else if (ptrace_ldt) {
238                 /* With PTRACE_LDT available, this is used as a flag only */
239                 ldt->entry_count = 1;
240                 goto out;
241         }
242
243         if (ldt_info.entry_number >= ldt->entry_count &&
244             ldt_info.entry_number >= LDT_DIRECT_ENTRIES) {
245                 for (i=ldt->entry_count/LDT_ENTRIES_PER_PAGE;
246                      i*LDT_ENTRIES_PER_PAGE <= ldt_info.entry_number;
247                      i++) {
248                         if (i == 0)
249                                 memcpy(&entry0, ldt->u.entries,
250                                        sizeof(entry0));
251                         ldt->u.pages[i] = (struct ldt_entry *)
252                                 __get_free_page(GFP_KERNEL|__GFP_ZERO);
253                         if (!ldt->u.pages[i]) {
254                                 err = -ENOMEM;
255                                 /* Undo the change in host */
256                                 memset(&ldt_info, 0, sizeof(ldt_info));
257                                 write_ldt_entry(mm_idp, 1, &ldt_info, &addr, 1);
258                                 goto out_unlock;
259                         }
260                         if (i == 0) {
261                                 memcpy(ldt->u.pages[0], &entry0,
262                                        sizeof(entry0));
263                                 memcpy(ldt->u.pages[0]+1, ldt->u.entries+1,
264                                        sizeof(entry0)*(LDT_DIRECT_ENTRIES-1));
265                         }
266                         ldt->entry_count = (i + 1) * LDT_ENTRIES_PER_PAGE;
267                 }
268         }
269         if (ldt->entry_count <= ldt_info.entry_number)
270                 ldt->entry_count = ldt_info.entry_number + 1;
271
272         if (ldt->entry_count <= LDT_DIRECT_ENTRIES)
273                 ldt_p = ldt->u.entries + ldt_info.entry_number;
274         else
275                 ldt_p = ldt->u.pages[ldt_info.entry_number/LDT_ENTRIES_PER_PAGE] +
276                         ldt_info.entry_number%LDT_ENTRIES_PER_PAGE;
277
278         if (ldt_info.base_addr == 0 && ldt_info.limit == 0 &&
279            (func == 1 || LDT_empty(&ldt_info))) {
280                 ldt_p->a = 0;
281                 ldt_p->b = 0;
282         }
283         else{
284                 if (func == 1)
285                         ldt_info.useable = 0;
286                 ldt_p->a = LDT_entry_a(&ldt_info);
287                 ldt_p->b = LDT_entry_b(&ldt_info);
288         }
289         err = 0;
290
291 out_unlock:
292         mutex_unlock(&ldt->lock);
293 out:
294         return err;
295 }
296
297 static long do_modify_ldt_skas(int func, void __user *ptr,
298                                unsigned long bytecount)
299 {
300         int ret = -ENOSYS;
301
302         switch (func) {
303                 case 0:
304                         ret = read_ldt(ptr, bytecount);
305                         break;
306                 case 1:
307                 case 0x11:
308                         ret = write_ldt(ptr, bytecount, func);
309                         break;
310                 case 2:
311                         ret = read_default_ldt(ptr, bytecount);
312                         break;
313         }
314         return ret;
315 }
316
317 static DEFINE_SPINLOCK(host_ldt_lock);
318 static short dummy_list[9] = {0, -1};
319 static short * host_ldt_entries = NULL;
320
321 static void ldt_get_host_info(void)
322 {
323         long ret;
324         struct ldt_entry * ldt;
325         short *tmp;
326         int i, size, k, order;
327
328         spin_lock(&host_ldt_lock);
329
330         if (host_ldt_entries != NULL) {
331                 spin_unlock(&host_ldt_lock);
332                 return;
333         }
334         host_ldt_entries = dummy_list+1;
335
336         spin_unlock(&host_ldt_lock);
337
338         for (i = LDT_PAGES_MAX-1, order=0; i; i>>=1, order++)
339                 ;
340
341         ldt = (struct ldt_entry *)
342               __get_free_pages(GFP_KERNEL|__GFP_ZERO, order);
343         if (ldt == NULL) {
344                 printk(KERN_ERR "ldt_get_host_info: couldn't allocate buffer "
345                        "for host ldt\n");
346                 return;
347         }
348
349         ret = modify_ldt(0, ldt, (1<<order)*PAGE_SIZE);
350         if (ret < 0) {
351                 printk(KERN_ERR "ldt_get_host_info: couldn't read host ldt\n");
352                 goto out_free;
353         }
354         if (ret == 0) {
355                 /* default_ldt is active, simply write an empty entry 0 */
356                 host_ldt_entries = dummy_list;
357                 goto out_free;
358         }
359
360         for (i=0, size=0; i<ret/LDT_ENTRY_SIZE; i++) {
361                 if (ldt[i].a != 0 || ldt[i].b != 0)
362                         size++;
363         }
364
365         if (size < ARRAY_SIZE(dummy_list))
366                 host_ldt_entries = dummy_list;
367         else {
368                 size = (size + 1) * sizeof(dummy_list[0]);
369                 tmp = kmalloc(size, GFP_KERNEL);
370                 if (tmp == NULL) {
371                         printk(KERN_ERR "ldt_get_host_info: couldn't allocate "
372                                "host ldt list\n");
373                         goto out_free;
374                 }
375                 host_ldt_entries = tmp;
376         }
377
378         for (i=0, k=0; i<ret/LDT_ENTRY_SIZE; i++) {
379                 if (ldt[i].a != 0 || ldt[i].b != 0)
380                         host_ldt_entries[k++] = i;
381         }
382         host_ldt_entries[k] = -1;
383
384 out_free:
385         free_pages((unsigned long)ldt, order);
386 }
387
388 long init_new_ldt(struct mm_context *new_mm, struct mm_context *from_mm)
389 {
390         struct user_desc desc;
391         short * num_p;
392         int i;
393         long page, err=0;
394         void *addr = NULL;
395         struct proc_mm_op copy;
396
397
398         if (!ptrace_ldt)
399                 mutex_init(&new_mm->ldt.lock);
400
401         if (!from_mm) {
402                 memset(&desc, 0, sizeof(desc));
403                 /*
404                  * We have to initialize a clean ldt.
405                  */
406                 if (proc_mm) {
407                         /*
408                          * If the new mm was created using proc_mm, host's
409                          * default-ldt currently is assigned, which normally
410                          * contains the call-gates for lcall7 and lcall27.
411                          * To remove these gates, we simply write an empty
412                          * entry as number 0 to the host.
413                          */
414                         err = write_ldt_entry(&new_mm->id, 1, &desc, &addr, 1);
415                 }
416                 else{
417                         /*
418                          * Now we try to retrieve info about the ldt, we
419                          * inherited from the host. All ldt-entries found
420                          * will be reset in the following loop
421                          */
422                         ldt_get_host_info();
423                         for (num_p=host_ldt_entries; *num_p != -1; num_p++) {
424                                 desc.entry_number = *num_p;
425                                 err = write_ldt_entry(&new_mm->id, 1, &desc,
426                                                       &addr, *(num_p + 1) == -1);
427                                 if (err)
428                                         break;
429                         }
430                 }
431                 new_mm->ldt.entry_count = 0;
432
433                 goto out;
434         }
435
436         if (proc_mm) {
437                 /*
438                  * We have a valid from_mm, so we now have to copy the LDT of
439                  * from_mm to new_mm, because using proc_mm an new mm with
440                  * an empty/default LDT was created in new_mm()
441                  */
442                 copy = ((struct proc_mm_op) { .op       = MM_COPY_SEGMENTS,
443                                               .u        =
444                                               { .copy_segments =
445                                                         from_mm->id.u.mm_fd } } );
446                 i = os_write_file(new_mm->id.u.mm_fd, &copy, sizeof(copy));
447                 if (i != sizeof(copy))
448                         printk(KERN_ERR "new_mm : /proc/mm copy_segments "
449                                "failed, err = %d\n", -i);
450         }
451
452         if (!ptrace_ldt) {
453                 /*
454                  * Our local LDT is used to supply the data for
455                  * modify_ldt(READLDT), if PTRACE_LDT isn't available,
456                  * i.e., we have to use the stub for modify_ldt, which
457                  * can't handle the big read buffer of up to 64kB.
458                  */
459                 mutex_lock(&from_mm->ldt.lock);
460                 if (from_mm->ldt.entry_count <= LDT_DIRECT_ENTRIES)
461                         memcpy(new_mm->ldt.u.entries, from_mm->ldt.u.entries,
462                                sizeof(new_mm->ldt.u.entries));
463                 else {
464                         i = from_mm->ldt.entry_count / LDT_ENTRIES_PER_PAGE;
465                         while (i-->0) {
466                                 page = __get_free_page(GFP_KERNEL|__GFP_ZERO);
467                                 if (!page) {
468                                         err = -ENOMEM;
469                                         break;
470                                 }
471                                 new_mm->ldt.u.pages[i] =
472                                         (struct ldt_entry *) page;
473                                 memcpy(new_mm->ldt.u.pages[i],
474                                        from_mm->ldt.u.pages[i], PAGE_SIZE);
475                         }
476                 }
477                 new_mm->ldt.entry_count = from_mm->ldt.entry_count;
478                 mutex_unlock(&from_mm->ldt.lock);
479         }
480
481     out:
482         return err;
483 }
484
485
486 void free_ldt(struct mm_context *mm)
487 {
488         int i;
489
490         if (!ptrace_ldt && mm->ldt.entry_count > LDT_DIRECT_ENTRIES) {
491                 i = mm->ldt.entry_count / LDT_ENTRIES_PER_PAGE;
492                 while (i-- > 0)
493                         free_page((long) mm->ldt.u.pages[i]);
494         }
495         mm->ldt.entry_count = 0;
496 }
497
498 int sys_modify_ldt(int func, void __user *ptr, unsigned long bytecount)
499 {
500         return do_modify_ldt_skas(func, ptr, bytecount);
501 }