Merge branch 'for-linus' of master.kernel.org:/pub/scm/linux/kernel/git/roland/infiniband
[pandora-kernel.git] / arch / um / sys-i386 / ldt.c
1 /*
2  * Copyright (C) 2001, 2002 Jeff Dike (jdike@karaya.com)
3  * Licensed under the GPL
4  */
5
6 #include "linux/stddef.h"
7 #include "linux/config.h"
8 #include "linux/sched.h"
9 #include "linux/slab.h"
10 #include "linux/types.h"
11 #include "linux/errno.h"
12 #include "asm/uaccess.h"
13 #include "asm/smp.h"
14 #include "asm/ldt.h"
15 #include "asm/unistd.h"
16 #include "choose-mode.h"
17 #include "kern.h"
18 #include "mode_kern.h"
19
20 extern int modify_ldt(int func, void *ptr, unsigned long bytecount);
21
22 #ifdef CONFIG_MODE_TT
23
24 static long do_modify_ldt_tt(int func, void __user *ptr,
25                               unsigned long bytecount)
26 {
27         struct user_desc info;
28         int res = 0;
29         void *buf = NULL;
30         void *p = NULL; /* What we pass to host. */
31
32         switch(func){
33         case 1:
34         case 0x11: /* write_ldt */
35                 /* Do this check now to avoid overflows. */
36                 if (bytecount != sizeof(struct user_desc)) {
37                         res = -EINVAL;
38                         goto out;
39                 }
40
41                 if(copy_from_user(&info, ptr, sizeof(info))) {
42                         res = -EFAULT;
43                         goto out;
44                 }
45
46                 p = &info;
47                 break;
48         case 0:
49         case 2: /* read_ldt */
50
51                 /* The use of info avoids kmalloc on the write case, not on the
52                  * read one. */
53                 buf = kmalloc(bytecount, GFP_KERNEL);
54                 if (!buf) {
55                         res = -ENOMEM;
56                         goto out;
57                 }
58                 p = buf;
59                 break;
60         default:
61                 res = -ENOSYS;
62                 goto out;
63         }
64
65         res = modify_ldt(func, p, bytecount);
66         if(res < 0)
67                 goto out;
68
69         switch(func){
70         case 0:
71         case 2:
72                 /* Modify_ldt was for reading and returned the number of read
73                  * bytes.*/
74                 if(copy_to_user(ptr, p, res))
75                         res = -EFAULT;
76                 break;
77         }
78
79 out:
80         kfree(buf);
81         return res;
82 }
83
84 #endif
85
86 #ifdef CONFIG_MODE_SKAS
87
88 #include "skas.h"
89 #include "skas_ptrace.h"
90 #include "asm/mmu_context.h"
91
92 long write_ldt_entry(struct mm_id * mm_idp, int func, struct user_desc * desc,
93                      void **addr, int done)
94 {
95         long res;
96
97         if(proc_mm){
98                 /* This is a special handling for the case, that the mm to
99                  * modify isn't current->active_mm.
100                  * If this is called directly by modify_ldt,
101                  *     (current->active_mm->context.skas.u == mm_idp)
102                  * will be true. So no call to switch_mm_skas(mm_idp) is done.
103                  * If this is called in case of init_new_ldt or PTRACE_LDT,
104                  * mm_idp won't belong to current->active_mm, but child->mm.
105                  * So we need to switch child's mm into our userspace, then
106                  * later switch back.
107                  *
108                  * Note: I'm unshure: should interrupts be disabled here?
109                  */
110                 if(!current->active_mm || current->active_mm == &init_mm ||
111                    mm_idp != &current->active_mm->context.skas.id)
112                         switch_mm_skas(mm_idp);
113         }
114
115         if(ptrace_ldt) {
116                 struct ptrace_ldt ldt_op = (struct ptrace_ldt) {
117                         .func = func,
118                         .ptr = desc,
119                         .bytecount = sizeof(*desc)};
120                 u32 cpu;
121                 int pid;
122
123                 if(!proc_mm)
124                         pid = mm_idp->u.pid;
125                 else {
126                         cpu = get_cpu();
127                         pid = userspace_pid[cpu];
128                 }
129
130                 res = ptrace(PTRACE_LDT, pid, 0, (unsigned long) &ldt_op);
131                 if(res)
132                         res = errno;
133
134                 if(proc_mm)
135                         put_cpu();
136         }
137         else {
138                 void *stub_addr;
139                 res = syscall_stub_data(mm_idp, (unsigned long *)desc,
140                                         (sizeof(*desc) + sizeof(long) - 1) &
141                                             ~(sizeof(long) - 1),
142                                         addr, &stub_addr);
143                 if(!res){
144                         unsigned long args[] = { func,
145                                                  (unsigned long)stub_addr,
146                                                  sizeof(*desc),
147                                                  0, 0, 0 };
148                         res = run_syscall_stub(mm_idp, __NR_modify_ldt, args,
149                                                0, addr, done);
150                 }
151         }
152
153         if(proc_mm){
154                 /* This is the second part of special handling, that makes
155                  * PTRACE_LDT possible to implement.
156                  */
157                 if(current->active_mm && current->active_mm != &init_mm &&
158                    mm_idp != &current->active_mm->context.skas.id)
159                         switch_mm_skas(&current->active_mm->context.skas.id);
160         }
161
162         return res;
163 }
164
165 static long read_ldt_from_host(void __user * ptr, unsigned long bytecount)
166 {
167         int res, n;
168         struct ptrace_ldt ptrace_ldt = (struct ptrace_ldt) {
169                         .func = 0,
170                         .bytecount = bytecount,
171                         .ptr = (void *)kmalloc(bytecount, GFP_KERNEL)};
172         u32 cpu;
173
174         if(ptrace_ldt.ptr == NULL)
175                 return -ENOMEM;
176
177         /* This is called from sys_modify_ldt only, so userspace_pid gives
178          * us the right number
179          */
180
181         cpu = get_cpu();
182         res = ptrace(PTRACE_LDT, userspace_pid[cpu], 0,
183                      (unsigned long) &ptrace_ldt);
184         put_cpu();
185         if(res < 0)
186                 goto out;
187
188         n = copy_to_user(ptr, ptrace_ldt.ptr, res);
189         if(n != 0)
190                 res = -EFAULT;
191
192   out:
193         kfree(ptrace_ldt.ptr);
194
195         return res;
196 }
197
198 /*
199  * In skas mode, we hold our own ldt data in UML.
200  * Thus, the code implementing sys_modify_ldt_skas
201  * is very similar to (and mostly stolen from) sys_modify_ldt
202  * for arch/i386/kernel/ldt.c
203  * The routines copied and modified in part are:
204  * - read_ldt
205  * - read_default_ldt
206  * - write_ldt
207  * - sys_modify_ldt_skas
208  */
209
210 static int read_ldt(void __user * ptr, unsigned long bytecount)
211 {
212         int i, err = 0;
213         unsigned long size;
214         uml_ldt_t * ldt = &current->mm->context.skas.ldt;
215
216         if(!ldt->entry_count)
217                 goto out;
218         if(bytecount > LDT_ENTRY_SIZE*LDT_ENTRIES)
219                 bytecount = LDT_ENTRY_SIZE*LDT_ENTRIES;
220         err = bytecount;
221
222         if(ptrace_ldt){
223                 return read_ldt_from_host(ptr, bytecount);
224         }
225
226         down(&ldt->semaphore);
227         if(ldt->entry_count <= LDT_DIRECT_ENTRIES){
228                 size = LDT_ENTRY_SIZE*LDT_DIRECT_ENTRIES;
229                 if(size > bytecount)
230                         size = bytecount;
231                 if(copy_to_user(ptr, ldt->entries, size))
232                         err = -EFAULT;
233                 bytecount -= size;
234                 ptr += size;
235         }
236         else {
237                 for(i=0; i<ldt->entry_count/LDT_ENTRIES_PER_PAGE && bytecount;
238                          i++){
239                         size = PAGE_SIZE;
240                         if(size > bytecount)
241                                 size = bytecount;
242                         if(copy_to_user(ptr, ldt->pages[i], size)){
243                                 err = -EFAULT;
244                                 break;
245                         }
246                         bytecount -= size;
247                         ptr += size;
248                 }
249         }
250         up(&ldt->semaphore);
251
252         if(bytecount == 0 || err == -EFAULT)
253                 goto out;
254
255         if(clear_user(ptr, bytecount))
256                 err = -EFAULT;
257
258 out:
259         return err;
260 }
261
262 static int read_default_ldt(void __user * ptr, unsigned long bytecount)
263 {
264         int err;
265
266         if(bytecount > 5*LDT_ENTRY_SIZE)
267                 bytecount = 5*LDT_ENTRY_SIZE;
268
269         err = bytecount;
270         /* UML doesn't support lcall7 and lcall27.
271          * So, we don't really have a default ldt, but emulate
272          * an empty ldt of common host default ldt size.
273          */
274         if(clear_user(ptr, bytecount))
275                 err = -EFAULT;
276
277         return err;
278 }
279
280 static int write_ldt(void __user * ptr, unsigned long bytecount, int func)
281 {
282         uml_ldt_t * ldt = &current->mm->context.skas.ldt;
283         struct mm_id * mm_idp = &current->mm->context.skas.id;
284         int i, err;
285         struct user_desc ldt_info;
286         struct ldt_entry entry0, *ldt_p;
287         void *addr = NULL;
288
289         err = -EINVAL;
290         if(bytecount != sizeof(ldt_info))
291                 goto out;
292         err = -EFAULT;
293         if(copy_from_user(&ldt_info, ptr, sizeof(ldt_info)))
294                 goto out;
295
296         err = -EINVAL;
297         if(ldt_info.entry_number >= LDT_ENTRIES)
298                 goto out;
299         if(ldt_info.contents == 3){
300                 if (func == 1)
301                         goto out;
302                 if (ldt_info.seg_not_present == 0)
303                         goto out;
304         }
305
306         if(!ptrace_ldt)
307                 down(&ldt->semaphore);
308
309         err = write_ldt_entry(mm_idp, func, &ldt_info, &addr, 1);
310         if(err)
311                 goto out_unlock;
312         else if(ptrace_ldt) {
313         /* With PTRACE_LDT available, this is used as a flag only */
314                 ldt->entry_count = 1;
315                 goto out;
316         }
317
318         if(ldt_info.entry_number >= ldt->entry_count &&
319            ldt_info.entry_number >= LDT_DIRECT_ENTRIES){
320                 for(i=ldt->entry_count/LDT_ENTRIES_PER_PAGE;
321                     i*LDT_ENTRIES_PER_PAGE <= ldt_info.entry_number;
322                     i++){
323                         if(i == 0)
324                                 memcpy(&entry0, ldt->entries, sizeof(entry0));
325                         ldt->pages[i] = (struct ldt_entry *)
326                                         __get_free_page(GFP_KERNEL|__GFP_ZERO);
327                         if(!ldt->pages[i]){
328                                 err = -ENOMEM;
329                                 /* Undo the change in host */
330                                 memset(&ldt_info, 0, sizeof(ldt_info));
331                                 write_ldt_entry(mm_idp, 1, &ldt_info, &addr, 1);
332                                 goto out_unlock;
333                         }
334                         if(i == 0) {
335                                 memcpy(ldt->pages[0], &entry0, sizeof(entry0));
336                                 memcpy(ldt->pages[0]+1, ldt->entries+1,
337                                        sizeof(entry0)*(LDT_DIRECT_ENTRIES-1));
338                         }
339                         ldt->entry_count = (i + 1) * LDT_ENTRIES_PER_PAGE;
340                 }
341         }
342         if(ldt->entry_count <= ldt_info.entry_number)
343                 ldt->entry_count = ldt_info.entry_number + 1;
344
345         if(ldt->entry_count <= LDT_DIRECT_ENTRIES)
346                 ldt_p = ldt->entries + ldt_info.entry_number;
347         else
348                 ldt_p = ldt->pages[ldt_info.entry_number/LDT_ENTRIES_PER_PAGE] +
349                         ldt_info.entry_number%LDT_ENTRIES_PER_PAGE;
350
351         if(ldt_info.base_addr == 0 && ldt_info.limit == 0 &&
352            (func == 1 || LDT_empty(&ldt_info))){
353                 ldt_p->a = 0;
354                 ldt_p->b = 0;
355         }
356         else{
357                 if (func == 1)
358                         ldt_info.useable = 0;
359                 ldt_p->a = LDT_entry_a(&ldt_info);
360                 ldt_p->b = LDT_entry_b(&ldt_info);
361         }
362         err = 0;
363
364 out_unlock:
365         up(&ldt->semaphore);
366 out:
367         return err;
368 }
369
370 static long do_modify_ldt_skas(int func, void __user *ptr,
371                                unsigned long bytecount)
372 {
373         int ret = -ENOSYS;
374
375         switch (func) {
376                 case 0:
377                         ret = read_ldt(ptr, bytecount);
378                         break;
379                 case 1:
380                 case 0x11:
381                         ret = write_ldt(ptr, bytecount, func);
382                         break;
383                 case 2:
384                         ret = read_default_ldt(ptr, bytecount);
385                         break;
386         }
387         return ret;
388 }
389
390 short dummy_list[9] = {0, -1};
391 short * host_ldt_entries = NULL;
392
393 void ldt_get_host_info(void)
394 {
395         long ret;
396         struct ldt_entry * ldt;
397         int i, size, k, order;
398
399         host_ldt_entries = dummy_list+1;
400
401         for(i = LDT_PAGES_MAX-1, order=0; i; i>>=1, order++);
402
403         ldt = (struct ldt_entry *)
404               __get_free_pages(GFP_KERNEL|__GFP_ZERO, order);
405         if(ldt == NULL) {
406                 printk("ldt_get_host_info: couldn't allocate buffer for host ldt\n");
407                 return;
408         }
409
410         ret = modify_ldt(0, ldt, (1<<order)*PAGE_SIZE);
411         if(ret < 0) {
412                 printk("ldt_get_host_info: couldn't read host ldt\n");
413                 goto out_free;
414         }
415         if(ret == 0) {
416                 /* default_ldt is active, simply write an empty entry 0 */
417                 host_ldt_entries = dummy_list;
418                 goto out_free;
419         }
420
421         for(i=0, size=0; i<ret/LDT_ENTRY_SIZE; i++){
422                 if(ldt[i].a != 0 || ldt[i].b != 0)
423                         size++;
424         }
425
426         if(size < sizeof(dummy_list)/sizeof(dummy_list[0])) {
427                 host_ldt_entries = dummy_list;
428         }
429         else {
430                 size = (size + 1) * sizeof(dummy_list[0]);
431                 host_ldt_entries = (short *)kmalloc(size, GFP_KERNEL);
432                 if(host_ldt_entries == NULL) {
433                         printk("ldt_get_host_info: couldn't allocate host ldt list\n");
434                         goto out_free;
435                 }
436         }
437
438         for(i=0, k=0; i<ret/LDT_ENTRY_SIZE; i++){
439                 if(ldt[i].a != 0 || ldt[i].b != 0) {
440                         host_ldt_entries[k++] = i;
441                 }
442         }
443         host_ldt_entries[k] = -1;
444
445 out_free:
446         free_pages((unsigned long)ldt, order);
447 }
448
449 long init_new_ldt(struct mmu_context_skas * new_mm,
450                   struct mmu_context_skas * from_mm)
451 {
452         struct user_desc desc;
453         short * num_p;
454         int i;
455         long page, err=0;
456         void *addr = NULL;
457
458         memset(&desc, 0, sizeof(desc));
459
460         if(!ptrace_ldt)
461                 init_MUTEX(&new_mm->ldt.semaphore);
462
463         if(!from_mm){
464                 /*
465                  * We have to initialize a clean ldt.
466                  */
467                 if(proc_mm) {
468                         /*
469                          * If the new mm was created using proc_mm, host's
470                          * default-ldt currently is assigned, which normally
471                          * contains the call-gates for lcall7 and lcall27.
472                          * To remove these gates, we simply write an empty
473                          * entry as number 0 to the host.
474                          */
475                         err = write_ldt_entry(&new_mm->id, 1, &desc,
476                                               &addr, 1);
477                 }
478                 else{
479                         /*
480                          * Now we try to retrieve info about the ldt, we
481                          * inherited from the host. All ldt-entries found
482                          * will be reset in the following loop
483                          */
484                         if(host_ldt_entries == NULL)
485                                 ldt_get_host_info();
486                         for(num_p=host_ldt_entries; *num_p != -1; num_p++){
487                                 desc.entry_number = *num_p;
488                                 err = write_ldt_entry(&new_mm->id, 1, &desc,
489                                                       &addr, *(num_p + 1) == -1);
490                                 if(err)
491                                         break;
492                         }
493                 }
494                 new_mm->ldt.entry_count = 0;
495         }
496         else if (!ptrace_ldt) {
497                 /* Our local LDT is used to supply the data for
498                  * modify_ldt(READLDT), if PTRACE_LDT isn't available,
499                  * i.e., we have to use the stub for modify_ldt, which
500                  * can't handle the big read buffer of up to 64kB.
501                  */
502                 down(&from_mm->ldt.semaphore);
503                 if(from_mm->ldt.entry_count <= LDT_DIRECT_ENTRIES){
504                         memcpy(new_mm->ldt.entries, from_mm->ldt.entries,
505                                sizeof(new_mm->ldt.entries));
506                 }
507                 else{
508                         i = from_mm->ldt.entry_count / LDT_ENTRIES_PER_PAGE;
509                         while(i-->0){
510                                 page = __get_free_page(GFP_KERNEL|__GFP_ZERO);
511                                 if (!page){
512                                         err = -ENOMEM;
513                                         break;
514                                 }
515                                 new_mm->ldt.pages[i] = (struct ldt_entry*)page;
516                                 memcpy(new_mm->ldt.pages[i],
517                                        from_mm->ldt.pages[i], PAGE_SIZE);
518                         }
519                 }
520                 new_mm->ldt.entry_count = from_mm->ldt.entry_count;
521                 up(&from_mm->ldt.semaphore);
522         }
523
524         return err;
525 }
526
527
528 void free_ldt(struct mmu_context_skas * mm)
529 {
530         int i;
531
532         if(!ptrace_ldt && mm->ldt.entry_count > LDT_DIRECT_ENTRIES){
533                 i = mm->ldt.entry_count / LDT_ENTRIES_PER_PAGE;
534                 while(i-- > 0){
535                         free_page((long )mm->ldt.pages[i]);
536                 }
537         }
538         mm->ldt.entry_count = 0;
539 }
540 #endif
541
542 int sys_modify_ldt(int func, void __user *ptr, unsigned long bytecount)
543 {
544         return(CHOOSE_MODE_PROC(do_modify_ldt_tt, do_modify_ldt_skas, func,
545                                 ptr, bytecount));
546 }