Merge master.kernel.org:/pub/scm/linux/kernel/git/wim/linux-2.6-watchdog
[linux-2.6] / 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/sched.h"
7 #include "linux/slab.h"
8 #include "linux/types.h"
9 #include "linux/errno.h"
10 #include "linux/spinlock.h"
11 #include "asm/uaccess.h"
12 #include "asm/smp.h"
13 #include "asm/ldt.h"
14 #include "asm/unistd.h"
15 #include "choose-mode.h"
16 #include "kern.h"
17 #include "mode_kern.h"
18 #include "os.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 #include "proc_mm.h"
92
93 long write_ldt_entry(struct mm_id * mm_idp, int func, struct user_desc * desc,
94                      void **addr, int done)
95 {
96         long res;
97
98         if(proc_mm){
99                 /* This is a special handling for the case, that the mm to
100                  * modify isn't current->active_mm.
101                  * If this is called directly by modify_ldt,
102                  *     (current->active_mm->context.skas.u == mm_idp)
103                  * will be true. So no call to switch_mm_skas(mm_idp) is done.
104                  * If this is called in case of init_new_ldt or PTRACE_LDT,
105                  * mm_idp won't belong to current->active_mm, but child->mm.
106                  * So we need to switch child's mm into our userspace, then
107                  * later switch back.
108                  *
109                  * Note: I'm unsure: should interrupts be disabled here?
110                  */
111                 if(!current->active_mm || current->active_mm == &init_mm ||
112                    mm_idp != &current->active_mm->context.skas.id)
113                         switch_mm_skas(mm_idp);
114         }
115
116         if(ptrace_ldt) {
117                 struct ptrace_ldt ldt_op = (struct ptrace_ldt) {
118                         .func = func,
119                         .ptr = desc,
120                         .bytecount = sizeof(*desc)};
121                 u32 cpu;
122                 int pid;
123
124                 if(!proc_mm)
125                         pid = mm_idp->u.pid;
126                 else {
127                         cpu = get_cpu();
128                         pid = userspace_pid[cpu];
129                 }
130
131                 res = os_ptrace_ldt(pid, 0, (unsigned long) &ldt_op);
132
133                 if(proc_mm)
134                         put_cpu();
135         }
136         else {
137                 void *stub_addr;
138                 res = syscall_stub_data(mm_idp, (unsigned long *)desc,
139                                         (sizeof(*desc) + sizeof(long) - 1) &
140                                             ~(sizeof(long) - 1),
141                                         addr, &stub_addr);
142                 if(!res){
143                         unsigned long args[] = { func,
144                                                  (unsigned long)stub_addr,
145                                                  sizeof(*desc),
146                                                  0, 0, 0 };
147                         res = run_syscall_stub(mm_idp, __NR_modify_ldt, args,
148                                                0, addr, done);
149                 }
150         }
151
152         if(proc_mm){
153                 /* This is the second part of special handling, that makes
154                  * PTRACE_LDT possible to implement.
155                  */
156                 if(current->active_mm && current->active_mm != &init_mm &&
157                    mm_idp != &current->active_mm->context.skas.id)
158                         switch_mm_skas(&current->active_mm->context.skas.id);
159         }
160
161         return res;
162 }
163
164 static long read_ldt_from_host(void __user * ptr, unsigned long bytecount)
165 {
166         int res, n;
167         struct ptrace_ldt ptrace_ldt = (struct ptrace_ldt) {
168                         .func = 0,
169                         .bytecount = bytecount,
170                         .ptr = kmalloc(bytecount, GFP_KERNEL)};
171         u32 cpu;
172
173         if(ptrace_ldt.ptr == NULL)
174                 return -ENOMEM;
175
176         /* This is called from sys_modify_ldt only, so userspace_pid gives
177          * us the right number
178          */
179
180         cpu = get_cpu();
181         res = os_ptrace_ldt(userspace_pid[cpu], 0, (unsigned long) &ptrace_ldt);
182         put_cpu();
183         if(res < 0)
184                 goto out;
185
186         n = copy_to_user(ptr, ptrace_ldt.ptr, res);
187         if(n != 0)
188                 res = -EFAULT;
189
190   out:
191         kfree(ptrace_ldt.ptr);
192
193         return res;
194 }
195
196 /*
197  * In skas mode, we hold our own ldt data in UML.
198  * Thus, the code implementing sys_modify_ldt_skas
199  * is very similar to (and mostly stolen from) sys_modify_ldt
200  * for arch/i386/kernel/ldt.c
201  * The routines copied and modified in part are:
202  * - read_ldt
203  * - read_default_ldt
204  * - write_ldt
205  * - sys_modify_ldt_skas
206  */
207
208 static int read_ldt(void __user * ptr, unsigned long bytecount)
209 {
210         int i, err = 0;
211         unsigned long size;
212         uml_ldt_t * ldt = &current->mm->context.skas.ldt;
213
214         if(!ldt->entry_count)
215                 goto out;
216         if(bytecount > LDT_ENTRY_SIZE*LDT_ENTRIES)
217                 bytecount = LDT_ENTRY_SIZE*LDT_ENTRIES;
218         err = bytecount;
219
220         if(ptrace_ldt){
221                 return read_ldt_from_host(ptr, bytecount);
222         }
223
224         down(&ldt->semaphore);
225         if(ldt->entry_count <= LDT_DIRECT_ENTRIES){
226                 size = LDT_ENTRY_SIZE*LDT_DIRECT_ENTRIES;
227                 if(size > bytecount)
228                         size = bytecount;
229                 if(copy_to_user(ptr, ldt->u.entries, size))
230                         err = -EFAULT;
231                 bytecount -= size;
232                 ptr += size;
233         }
234         else {
235                 for(i=0; i<ldt->entry_count/LDT_ENTRIES_PER_PAGE && bytecount;
236                          i++){
237                         size = PAGE_SIZE;
238                         if(size > bytecount)
239                                 size = bytecount;
240                         if(copy_to_user(ptr, ldt->u.pages[i], size)){
241                                 err = -EFAULT;
242                                 break;
243                         }
244                         bytecount -= size;
245                         ptr += size;
246                 }
247         }
248         up(&ldt->semaphore);
249
250         if(bytecount == 0 || err == -EFAULT)
251                 goto out;
252
253         if(clear_user(ptr, bytecount))
254                 err = -EFAULT;
255
256 out:
257         return err;
258 }
259
260 static int read_default_ldt(void __user * ptr, unsigned long bytecount)
261 {
262         int err;
263
264         if(bytecount > 5*LDT_ENTRY_SIZE)
265                 bytecount = 5*LDT_ENTRY_SIZE;
266
267         err = bytecount;
268         /* UML doesn't support lcall7 and lcall27.
269          * So, we don't really have a default ldt, but emulate
270          * an empty ldt of common host default ldt size.
271          */
272         if(clear_user(ptr, bytecount))
273                 err = -EFAULT;
274
275         return err;
276 }
277
278 static int write_ldt(void __user * ptr, unsigned long bytecount, int func)
279 {
280         uml_ldt_t * ldt = &current->mm->context.skas.ldt;
281         struct mm_id * mm_idp = &current->mm->context.skas.id;
282         int i, err;
283         struct user_desc ldt_info;
284         struct ldt_entry entry0, *ldt_p;
285         void *addr = NULL;
286
287         err = -EINVAL;
288         if(bytecount != sizeof(ldt_info))
289                 goto out;
290         err = -EFAULT;
291         if(copy_from_user(&ldt_info, ptr, sizeof(ldt_info)))
292                 goto out;
293
294         err = -EINVAL;
295         if(ldt_info.entry_number >= LDT_ENTRIES)
296                 goto out;
297         if(ldt_info.contents == 3){
298                 if (func == 1)
299                         goto out;
300                 if (ldt_info.seg_not_present == 0)
301                         goto out;
302         }
303
304         if(!ptrace_ldt)
305                 down(&ldt->semaphore);
306
307         err = write_ldt_entry(mm_idp, func, &ldt_info, &addr, 1);
308         if(err)
309                 goto out_unlock;
310         else if(ptrace_ldt) {
311         /* With PTRACE_LDT available, this is used as a flag only */
312                 ldt->entry_count = 1;
313                 goto out;
314         }
315
316         if(ldt_info.entry_number >= ldt->entry_count &&
317            ldt_info.entry_number >= LDT_DIRECT_ENTRIES){
318                 for(i=ldt->entry_count/LDT_ENTRIES_PER_PAGE;
319                     i*LDT_ENTRIES_PER_PAGE <= ldt_info.entry_number;
320                     i++){
321                         if(i == 0)
322                                 memcpy(&entry0, ldt->u.entries,
323                                        sizeof(entry0));
324                         ldt->u.pages[i] = (struct ldt_entry *)
325                                 __get_free_page(GFP_KERNEL|__GFP_ZERO);
326                         if(!ldt->u.pages[i]){
327                                 err = -ENOMEM;
328                                 /* Undo the change in host */
329                                 memset(&ldt_info, 0, sizeof(ldt_info));
330                                 write_ldt_entry(mm_idp, 1, &ldt_info, &addr, 1);
331                                 goto out_unlock;
332                         }
333                         if(i == 0) {
334                                 memcpy(ldt->u.pages[0], &entry0,
335                                        sizeof(entry0));
336                                 memcpy(ldt->u.pages[0]+1, ldt->u.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->u.entries + ldt_info.entry_number;
347         else
348                 ldt_p = ldt->u.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 static DEFINE_SPINLOCK(host_ldt_lock);
391 static short dummy_list[9] = {0, -1};
392 static short * host_ldt_entries = NULL;
393
394 static void ldt_get_host_info(void)
395 {
396         long ret;
397         struct ldt_entry * ldt;
398         short *tmp;
399         int i, size, k, order;
400
401         spin_lock(&host_ldt_lock);
402
403         if(host_ldt_entries != NULL){
404                 spin_unlock(&host_ldt_lock);
405                 return;
406         }
407         host_ldt_entries = dummy_list+1;
408
409         spin_unlock(&host_ldt_lock);
410
411         for(i = LDT_PAGES_MAX-1, order=0; i; i>>=1, order++);
412
413         ldt = (struct ldt_entry *)
414               __get_free_pages(GFP_KERNEL|__GFP_ZERO, order);
415         if(ldt == NULL) {
416                 printk("ldt_get_host_info: couldn't allocate buffer for host "
417                        "ldt\n");
418                 return;
419         }
420
421         ret = modify_ldt(0, ldt, (1<<order)*PAGE_SIZE);
422         if(ret < 0) {
423                 printk("ldt_get_host_info: couldn't read host ldt\n");
424                 goto out_free;
425         }
426         if(ret == 0) {
427                 /* default_ldt is active, simply write an empty entry 0 */
428                 host_ldt_entries = dummy_list;
429                 goto out_free;
430         }
431
432         for(i=0, size=0; i<ret/LDT_ENTRY_SIZE; i++){
433                 if(ldt[i].a != 0 || ldt[i].b != 0)
434                         size++;
435         }
436
437         if(size < ARRAY_SIZE(dummy_list))
438                 host_ldt_entries = dummy_list;
439         else {
440                 size = (size + 1) * sizeof(dummy_list[0]);
441                 tmp = kmalloc(size, GFP_KERNEL);
442                 if(tmp == NULL) {
443                         printk("ldt_get_host_info: couldn't allocate host ldt "
444                                "list\n");
445                         goto out_free;
446                 }
447                 host_ldt_entries = tmp;
448         }
449
450         for(i=0, k=0; i<ret/LDT_ENTRY_SIZE; i++){
451                 if(ldt[i].a != 0 || ldt[i].b != 0) {
452                         host_ldt_entries[k++] = i;
453                 }
454         }
455         host_ldt_entries[k] = -1;
456
457 out_free:
458         free_pages((unsigned long)ldt, order);
459 }
460
461 long init_new_ldt(struct mmu_context_skas * new_mm,
462                   struct mmu_context_skas * from_mm)
463 {
464         struct user_desc desc;
465         short * num_p;
466         int i;
467         long page, err=0;
468         void *addr = NULL;
469         struct proc_mm_op copy;
470
471
472         if(!ptrace_ldt)
473                 init_MUTEX(&new_mm->ldt.semaphore);
474
475         if(!from_mm){
476                 memset(&desc, 0, sizeof(desc));
477                 /*
478                  * We have to initialize a clean ldt.
479                  */
480                 if(proc_mm) {
481                         /*
482                          * If the new mm was created using proc_mm, host's
483                          * default-ldt currently is assigned, which normally
484                          * contains the call-gates for lcall7 and lcall27.
485                          * To remove these gates, we simply write an empty
486                          * entry as number 0 to the host.
487                          */
488                         err = write_ldt_entry(&new_mm->id, 1, &desc,
489                                               &addr, 1);
490                 }
491                 else{
492                         /*
493                          * Now we try to retrieve info about the ldt, we
494                          * inherited from the host. All ldt-entries found
495                          * will be reset in the following loop
496                          */
497                         ldt_get_host_info();
498                         for(num_p=host_ldt_entries; *num_p != -1; num_p++){
499                                 desc.entry_number = *num_p;
500                                 err = write_ldt_entry(&new_mm->id, 1, &desc,
501                                                       &addr, *(num_p + 1) == -1);
502                                 if(err)
503                                         break;
504                         }
505                 }
506                 new_mm->ldt.entry_count = 0;
507
508                 goto out;
509         }
510
511         if(proc_mm){
512                 /* We have a valid from_mm, so we now have to copy the LDT of
513                  * from_mm to new_mm, because using proc_mm an new mm with
514                  * an empty/default LDT was created in new_mm()
515                  */
516                 copy = ((struct proc_mm_op) { .op       = MM_COPY_SEGMENTS,
517                                               .u        =
518                                               { .copy_segments =
519                                                         from_mm->id.u.mm_fd } } );
520                 i = os_write_file(new_mm->id.u.mm_fd, &copy, sizeof(copy));
521                 if(i != sizeof(copy))
522                         printk("new_mm : /proc/mm copy_segments failed, "
523                                "err = %d\n", -i);
524         }
525
526         if(!ptrace_ldt) {
527                 /* Our local LDT is used to supply the data for
528                  * modify_ldt(READLDT), if PTRACE_LDT isn't available,
529                  * i.e., we have to use the stub for modify_ldt, which
530                  * can't handle the big read buffer of up to 64kB.
531                  */
532                 down(&from_mm->ldt.semaphore);
533                 if(from_mm->ldt.entry_count <= LDT_DIRECT_ENTRIES){
534                         memcpy(new_mm->ldt.u.entries, from_mm->ldt.u.entries,
535                                sizeof(new_mm->ldt.u.entries));
536                 }
537                 else{
538                         i = from_mm->ldt.entry_count / LDT_ENTRIES_PER_PAGE;
539                         while(i-->0){
540                                 page = __get_free_page(GFP_KERNEL|__GFP_ZERO);
541                                 if (!page){
542                                         err = -ENOMEM;
543                                         break;
544                                 }
545                                 new_mm->ldt.u.pages[i] =
546                                         (struct ldt_entry *) page;
547                                 memcpy(new_mm->ldt.u.pages[i],
548                                        from_mm->ldt.u.pages[i], PAGE_SIZE);
549                         }
550                 }
551                 new_mm->ldt.entry_count = from_mm->ldt.entry_count;
552                 up(&from_mm->ldt.semaphore);
553         }
554
555     out:
556         return err;
557 }
558
559
560 void free_ldt(struct mmu_context_skas * mm)
561 {
562         int i;
563
564         if(!ptrace_ldt && mm->ldt.entry_count > LDT_DIRECT_ENTRIES){
565                 i = mm->ldt.entry_count / LDT_ENTRIES_PER_PAGE;
566                 while(i-- > 0){
567                         free_page((long )mm->ldt.u.pages[i]);
568                 }
569         }
570         mm->ldt.entry_count = 0;
571 }
572 #endif
573
574 int sys_modify_ldt(int func, void __user *ptr, unsigned long bytecount)
575 {
576         return CHOOSE_MODE_PROC(do_modify_ldt_tt, do_modify_ldt_skas, func,
577                                 ptr, bytecount);
578 }