uml: tidy pgtable.h
[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 "asm/unistd.h"
8 #include "os.h"
9 #include "proc_mm.h"
10 #include "skas.h"
11 #include "skas_ptrace.h"
12 #include "sysdep/tls.h"
13
14 extern int modify_ldt(int func, void *ptr, unsigned long bytecount);
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
21         if (proc_mm) {
22                 /*
23                  * This is a special handling for the case, that the mm to
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)
27                  * will be true. So no call to __switch_mm(mm_idp) is done.
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                  *
33                  * Note: I'm unsure: should interrupts be disabled here?
34                  */
35                 if (!current->active_mm || current->active_mm == &init_mm ||
36                     mm_idp != &current->active_mm->context.id)
37                         __switch_mm(mm_idp);
38         }
39
40         if (ptrace_ldt) {
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
48                 if (!proc_mm)
49                         pid = mm_idp->u.pid;
50                 else {
51                         cpu = get_cpu();
52                         pid = userspace_pid[cpu];
53                 }
54
55                 res = os_ptrace_ldt(pid, 0, (unsigned long) &ldt_op);
56
57                 if (proc_mm)
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);
66                 if (!res) {
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
76         if (proc_mm) {
77                 /*
78                  * This is the second part of special handling, that makes
79                  * PTRACE_LDT possible to implement.
80                  */
81                 if (current->active_mm && current->active_mm != &init_mm &&
82                     mm_idp != &current->active_mm->context.id)
83                         __switch_mm(&current->active_mm->context.id);
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,
95                         .ptr = kmalloc(bytecount, GFP_KERNEL)};
96         u32 cpu;
97
98         if (ptrace_ldt.ptr == NULL)
99                 return -ENOMEM;
100
101         /*
102          * This is called from sys_modify_ldt only, so userspace_pid gives
103          * us the right number
104          */
105
106         cpu = get_cpu();
107         res = os_ptrace_ldt(userspace_pid[cpu], 0, (unsigned long) &ptrace_ldt);
108         put_cpu();
109         if (res < 0)
110                 goto out;
111
112         n = copy_to_user(ptr, ptrace_ldt.ptr, res);
113         if (n != 0)
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;
138         uml_ldt_t * ldt = &current->mm->context.ldt;
139
140         if (!ldt->entry_count)
141                 goto out;
142         if (bytecount > LDT_ENTRY_SIZE*LDT_ENTRIES)
143                 bytecount = LDT_ENTRY_SIZE*LDT_ENTRIES;
144         err = bytecount;
145
146         if (ptrace_ldt)
147                 return read_ldt_from_host(ptr, bytecount);
148
149         down(&ldt->semaphore);
150         if (ldt->entry_count <= LDT_DIRECT_ENTRIES) {
151                 size = LDT_ENTRY_SIZE*LDT_DIRECT_ENTRIES;
152                 if (size > bytecount)
153                         size = bytecount;
154                 if (copy_to_user(ptr, ldt->u.entries, size))
155                         err = -EFAULT;
156                 bytecount -= size;
157                 ptr += size;
158         }
159         else {
160                 for (i=0; i<ldt->entry_count/LDT_ENTRIES_PER_PAGE && bytecount;
161                      i++) {
162                         size = PAGE_SIZE;
163                         if (size > bytecount)
164                                 size = bytecount;
165                         if (copy_to_user(ptr, ldt->u.pages[i], size)) {
166                                 err = -EFAULT;
167                                 break;
168                         }
169                         bytecount -= size;
170                         ptr += size;
171                 }
172         }
173         up(&ldt->semaphore);
174
175         if (bytecount == 0 || err == -EFAULT)
176                 goto out;
177
178         if (clear_user(ptr, bytecount))
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
189         if (bytecount > 5*LDT_ENTRY_SIZE)
190                 bytecount = 5*LDT_ENTRY_SIZE;
191
192         err = bytecount;
193         /*
194          * UML doesn't support lcall7 and lcall27.
195          * So, we don't really have a default ldt, but emulate
196          * an empty ldt of common host default ldt size.
197          */
198         if (clear_user(ptr, bytecount))
199                 err = -EFAULT;
200
201         return err;
202 }
203
204 static int write_ldt(void __user * ptr, unsigned long bytecount, int func)
205 {
206         uml_ldt_t * ldt = &current->mm->context.ldt;
207         struct mm_id * mm_idp = &current->mm->context.id;
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;
214         if (bytecount != sizeof(ldt_info))
215                 goto out;
216         err = -EFAULT;
217         if (copy_from_user(&ldt_info, ptr, sizeof(ldt_info)))
218                 goto out;
219
220         err = -EINVAL;
221         if (ldt_info.entry_number >= LDT_ENTRIES)
222                 goto out;
223         if (ldt_info.contents == 3) {
224                 if (func == 1)
225                         goto out;
226                 if (ldt_info.seg_not_present == 0)
227                         goto out;
228         }
229
230         if (!ptrace_ldt)
231                 down(&ldt->semaphore);
232
233         err = write_ldt_entry(mm_idp, func, &ldt_info, &addr, 1);
234         if (err)
235                 goto out_unlock;
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)
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);
252                         if (!ldt->u.pages[i]) {
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                         }
259                         if (i == 0) {
260                                 memcpy(ldt->u.pages[0], &entry0,
261                                        sizeof(entry0));
262                                 memcpy(ldt->u.pages[0]+1, ldt->u.entries+1,
263                                        sizeof(entry0)*(LDT_DIRECT_ENTRIES-1));
264                         }
265                         ldt->entry_count = (i + 1) * LDT_ENTRIES_PER_PAGE;
266                 }
267         }
268         if (ldt->entry_count <= ldt_info.entry_number)
269                 ldt->entry_count = ldt_info.entry_number + 1;
270
271         if (ldt->entry_count <= LDT_DIRECT_ENTRIES)
272                 ldt_p = ldt->u.entries + ldt_info.entry_number;
273         else
274                 ldt_p = ldt->u.pages[ldt_info.entry_number/LDT_ENTRIES_PER_PAGE] +
275                         ldt_info.entry_number%LDT_ENTRIES_PER_PAGE;
276
277         if (ldt_info.base_addr == 0 && ldt_info.limit == 0 &&
278            (func == 1 || LDT_empty(&ldt_info))) {
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
316 static DEFINE_SPINLOCK(host_ldt_lock);
317 static short dummy_list[9] = {0, -1};
318 static short * host_ldt_entries = NULL;
319
320 static void ldt_get_host_info(void)
321 {
322         long ret;
323         struct ldt_entry * ldt;
324         short *tmp;
325         int i, size, k, order;
326
327         spin_lock(&host_ldt_lock);
328
329         if (host_ldt_entries != NULL) {
330                 spin_unlock(&host_ldt_lock);
331                 return;
332         }
333         host_ldt_entries = dummy_list+1;
334
335         spin_unlock(&host_ldt_lock);
336
337         for (i = LDT_PAGES_MAX-1, order=0; i; i>>=1, order++)
338                 ;
339
340         ldt = (struct ldt_entry *)
341               __get_free_pages(GFP_KERNEL|__GFP_ZERO, order);
342         if (ldt == NULL) {
343                 printk(KERN_ERR "ldt_get_host_info: couldn't allocate buffer "
344                        "for host ldt\n");
345                 return;
346         }
347
348         ret = modify_ldt(0, ldt, (1<<order)*PAGE_SIZE);
349         if (ret < 0) {
350                 printk(KERN_ERR "ldt_get_host_info: couldn't read host ldt\n");
351                 goto out_free;
352         }
353         if (ret == 0) {
354                 /* default_ldt is active, simply write an empty entry 0 */
355                 host_ldt_entries = dummy_list;
356                 goto out_free;
357         }
358
359         for (i=0, size=0; i<ret/LDT_ENTRY_SIZE; i++) {
360                 if (ldt[i].a != 0 || ldt[i].b != 0)
361                         size++;
362         }
363
364         if (size < ARRAY_SIZE(dummy_list))
365                 host_ldt_entries = dummy_list;
366         else {
367                 size = (size + 1) * sizeof(dummy_list[0]);
368                 tmp = kmalloc(size, GFP_KERNEL);
369                 if (tmp == NULL) {
370                         printk(KERN_ERR "ldt_get_host_info: couldn't allocate "
371                                "host ldt list\n");
372                         goto out_free;
373                 }
374                 host_ldt_entries = tmp;
375         }
376
377         for (i=0, k=0; i<ret/LDT_ENTRY_SIZE; i++) {
378                 if (ldt[i].a != 0 || ldt[i].b != 0)
379                         host_ldt_entries[k++] = i;
380         }
381         host_ldt_entries[k] = -1;
382
383 out_free:
384         free_pages((unsigned long)ldt, order);
385 }
386
387 long init_new_ldt(struct mm_context *new_mm, struct mm_context *from_mm)
388 {
389         struct user_desc desc;
390         short * num_p;
391         int i;
392         long page, err=0;
393         void *addr = NULL;
394         struct proc_mm_op copy;
395
396
397         if (!ptrace_ldt)
398                 init_MUTEX(&new_mm->ldt.semaphore);
399
400         if (!from_mm) {
401                 memset(&desc, 0, sizeof(desc));
402                 /*
403                  * We have to initialize a clean ldt.
404                  */
405                 if (proc_mm) {
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                          */
413                         err = write_ldt_entry(&new_mm->id, 1, &desc, &addr, 1);
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                          */
421                         ldt_get_host_info();
422                         for (num_p=host_ldt_entries; *num_p != -1; num_p++) {
423                                 desc.entry_number = *num_p;
424                                 err = write_ldt_entry(&new_mm->id, 1, &desc,
425                                                       &addr, *(num_p + 1) == -1);
426                                 if (err)
427                                         break;
428                         }
429                 }
430                 new_mm->ldt.entry_count = 0;
431
432                 goto out;
433         }
434
435         if (proc_mm) {
436                 /*
437                  * We have a valid from_mm, so we now have to copy the LDT of
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 } } );
445                 i = os_write_file(new_mm->id.u.mm_fd, &copy, sizeof(copy));
446                 if (i != sizeof(copy))
447                         printk(KERN_ERR "new_mm : /proc/mm copy_segments "
448                                "failed, err = %d\n", -i);
449         }
450
451         if (!ptrace_ldt) {
452                 /*
453                  * Our local LDT is used to supply the data for
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);
459                 if (from_mm->ldt.entry_count <= LDT_DIRECT_ENTRIES)
460                         memcpy(new_mm->ldt.u.entries, from_mm->ldt.u.entries,
461                                sizeof(new_mm->ldt.u.entries));
462                 else {
463                         i = from_mm->ldt.entry_count / LDT_ENTRIES_PER_PAGE;
464                         while (i-->0) {
465                                 page = __get_free_page(GFP_KERNEL|__GFP_ZERO);
466                                 if (!page) {
467                                         err = -ENOMEM;
468                                         break;
469                                 }
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);
474                         }
475                 }
476                 new_mm->ldt.entry_count = from_mm->ldt.entry_count;
477                 up(&from_mm->ldt.semaphore);
478         }
479
480     out:
481         return err;
482 }
483
484
485 void free_ldt(struct mm_context *mm)
486 {
487         int i;
488
489         if (!ptrace_ldt && mm->ldt.entry_count > LDT_DIRECT_ENTRIES) {
490                 i = mm->ldt.entry_count / LDT_ENTRIES_PER_PAGE;
491                 while (i-- > 0)
492                         free_page((long) mm->ldt.u.pages[i]);
493         }
494         mm->ldt.entry_count = 0;
495 }
496
497 int sys_modify_ldt(int func, void __user *ptr, unsigned long bytecount)
498 {
499         return do_modify_ldt_skas(func, ptr, bytecount);
500 }