um: take arch/um/sys-x86 to arch/x86/um
[pohmelfs.git] / arch / x86 / um / ldt.c
blob3f2bf208d884ec6973d550f84a9f5d0625ae2e5f
1 /*
2 * Copyright (C) 2001 - 2007 Jeff Dike (jdike@{addtoit,linux.intel}.com)
3 * Licensed under the GPL
4 */
6 #include <linux/mm.h>
7 #include <linux/sched.h>
8 #include <linux/slab.h>
9 #include <asm/unistd.h>
10 #include "os.h"
11 #include "proc_mm.h"
12 #include "skas.h"
13 #include "skas_ptrace.h"
14 #include "sysdep/tls.h"
16 extern int modify_ldt(int func, void *ptr, unsigned long bytecount);
18 static long write_ldt_entry(struct mm_id *mm_idp, int func,
19 struct user_desc *desc, void **addr, int done)
21 long res;
23 if (proc_mm) {
25 * This is a special handling for the case, that the mm to
26 * modify isn't current->active_mm.
27 * If this is called directly by modify_ldt,
28 * (current->active_mm->context.skas.u == mm_idp)
29 * will be true. So no call to __switch_mm(mm_idp) is done.
30 * If this is called in case of init_new_ldt or PTRACE_LDT,
31 * mm_idp won't belong to current->active_mm, but child->mm.
32 * So we need to switch child's mm into our userspace, then
33 * later switch back.
35 * Note: I'm unsure: should interrupts be disabled here?
37 if (!current->active_mm || current->active_mm == &init_mm ||
38 mm_idp != &current->active_mm->context.id)
39 __switch_mm(mm_idp);
42 if (ptrace_ldt) {
43 struct ptrace_ldt ldt_op = (struct ptrace_ldt) {
44 .func = func,
45 .ptr = desc,
46 .bytecount = sizeof(*desc)};
47 u32 cpu;
48 int pid;
50 if (!proc_mm)
51 pid = mm_idp->u.pid;
52 else {
53 cpu = get_cpu();
54 pid = userspace_pid[cpu];
57 res = os_ptrace_ldt(pid, 0, (unsigned long) &ldt_op);
59 if (proc_mm)
60 put_cpu();
62 else {
63 void *stub_addr;
64 res = syscall_stub_data(mm_idp, (unsigned long *)desc,
65 (sizeof(*desc) + sizeof(long) - 1) &
66 ~(sizeof(long) - 1),
67 addr, &stub_addr);
68 if (!res) {
69 unsigned long args[] = { func,
70 (unsigned long)stub_addr,
71 sizeof(*desc),
72 0, 0, 0 };
73 res = run_syscall_stub(mm_idp, __NR_modify_ldt, args,
74 0, addr, done);
78 if (proc_mm) {
80 * This is the second part of special handling, that makes
81 * PTRACE_LDT possible to implement.
83 if (current->active_mm && current->active_mm != &init_mm &&
84 mm_idp != &current->active_mm->context.id)
85 __switch_mm(&current->active_mm->context.id);
88 return res;
91 static long read_ldt_from_host(void __user * ptr, unsigned long bytecount)
93 int res, n;
94 struct ptrace_ldt ptrace_ldt = (struct ptrace_ldt) {
95 .func = 0,
96 .bytecount = bytecount,
97 .ptr = kmalloc(bytecount, GFP_KERNEL)};
98 u32 cpu;
100 if (ptrace_ldt.ptr == NULL)
101 return -ENOMEM;
104 * This is called from sys_modify_ldt only, so userspace_pid gives
105 * us the right number
108 cpu = get_cpu();
109 res = os_ptrace_ldt(userspace_pid[cpu], 0, (unsigned long) &ptrace_ldt);
110 put_cpu();
111 if (res < 0)
112 goto out;
114 n = copy_to_user(ptr, ptrace_ldt.ptr, res);
115 if (n != 0)
116 res = -EFAULT;
118 out:
119 kfree(ptrace_ldt.ptr);
121 return res;
125 * In skas mode, we hold our own ldt data in UML.
126 * Thus, the code implementing sys_modify_ldt_skas
127 * is very similar to (and mostly stolen from) sys_modify_ldt
128 * for arch/i386/kernel/ldt.c
129 * The routines copied and modified in part are:
130 * - read_ldt
131 * - read_default_ldt
132 * - write_ldt
133 * - sys_modify_ldt_skas
136 static int read_ldt(void __user * ptr, unsigned long bytecount)
138 int i, err = 0;
139 unsigned long size;
140 uml_ldt_t * ldt = &current->mm->context.ldt;
142 if (!ldt->entry_count)
143 goto out;
144 if (bytecount > LDT_ENTRY_SIZE*LDT_ENTRIES)
145 bytecount = LDT_ENTRY_SIZE*LDT_ENTRIES;
146 err = bytecount;
148 if (ptrace_ldt)
149 return read_ldt_from_host(ptr, bytecount);
151 mutex_lock(&ldt->lock);
152 if (ldt->entry_count <= LDT_DIRECT_ENTRIES) {
153 size = LDT_ENTRY_SIZE*LDT_DIRECT_ENTRIES;
154 if (size > bytecount)
155 size = bytecount;
156 if (copy_to_user(ptr, ldt->u.entries, size))
157 err = -EFAULT;
158 bytecount -= size;
159 ptr += size;
161 else {
162 for (i=0; i<ldt->entry_count/LDT_ENTRIES_PER_PAGE && bytecount;
163 i++) {
164 size = PAGE_SIZE;
165 if (size > bytecount)
166 size = bytecount;
167 if (copy_to_user(ptr, ldt->u.pages[i], size)) {
168 err = -EFAULT;
169 break;
171 bytecount -= size;
172 ptr += size;
175 mutex_unlock(&ldt->lock);
177 if (bytecount == 0 || err == -EFAULT)
178 goto out;
180 if (clear_user(ptr, bytecount))
181 err = -EFAULT;
183 out:
184 return err;
187 static int read_default_ldt(void __user * ptr, unsigned long bytecount)
189 int err;
191 if (bytecount > 5*LDT_ENTRY_SIZE)
192 bytecount = 5*LDT_ENTRY_SIZE;
194 err = bytecount;
196 * UML doesn't support lcall7 and lcall27.
197 * So, we don't really have a default ldt, but emulate
198 * an empty ldt of common host default ldt size.
200 if (clear_user(ptr, bytecount))
201 err = -EFAULT;
203 return err;
206 static int write_ldt(void __user * ptr, unsigned long bytecount, int func)
208 uml_ldt_t * ldt = &current->mm->context.ldt;
209 struct mm_id * mm_idp = &current->mm->context.id;
210 int i, err;
211 struct user_desc ldt_info;
212 struct ldt_entry entry0, *ldt_p;
213 void *addr = NULL;
215 err = -EINVAL;
216 if (bytecount != sizeof(ldt_info))
217 goto out;
218 err = -EFAULT;
219 if (copy_from_user(&ldt_info, ptr, sizeof(ldt_info)))
220 goto out;
222 err = -EINVAL;
223 if (ldt_info.entry_number >= LDT_ENTRIES)
224 goto out;
225 if (ldt_info.contents == 3) {
226 if (func == 1)
227 goto out;
228 if (ldt_info.seg_not_present == 0)
229 goto out;
232 if (!ptrace_ldt)
233 mutex_lock(&ldt->lock);
235 err = write_ldt_entry(mm_idp, func, &ldt_info, &addr, 1);
236 if (err)
237 goto out_unlock;
238 else if (ptrace_ldt) {
239 /* With PTRACE_LDT available, this is used as a flag only */
240 ldt->entry_count = 1;
241 goto out;
244 if (ldt_info.entry_number >= ldt->entry_count &&
245 ldt_info.entry_number >= LDT_DIRECT_ENTRIES) {
246 for (i=ldt->entry_count/LDT_ENTRIES_PER_PAGE;
247 i*LDT_ENTRIES_PER_PAGE <= ldt_info.entry_number;
248 i++) {
249 if (i == 0)
250 memcpy(&entry0, ldt->u.entries,
251 sizeof(entry0));
252 ldt->u.pages[i] = (struct ldt_entry *)
253 __get_free_page(GFP_KERNEL|__GFP_ZERO);
254 if (!ldt->u.pages[i]) {
255 err = -ENOMEM;
256 /* Undo the change in host */
257 memset(&ldt_info, 0, sizeof(ldt_info));
258 write_ldt_entry(mm_idp, 1, &ldt_info, &addr, 1);
259 goto out_unlock;
261 if (i == 0) {
262 memcpy(ldt->u.pages[0], &entry0,
263 sizeof(entry0));
264 memcpy(ldt->u.pages[0]+1, ldt->u.entries+1,
265 sizeof(entry0)*(LDT_DIRECT_ENTRIES-1));
267 ldt->entry_count = (i + 1) * LDT_ENTRIES_PER_PAGE;
270 if (ldt->entry_count <= ldt_info.entry_number)
271 ldt->entry_count = ldt_info.entry_number + 1;
273 if (ldt->entry_count <= LDT_DIRECT_ENTRIES)
274 ldt_p = ldt->u.entries + ldt_info.entry_number;
275 else
276 ldt_p = ldt->u.pages[ldt_info.entry_number/LDT_ENTRIES_PER_PAGE] +
277 ldt_info.entry_number%LDT_ENTRIES_PER_PAGE;
279 if (ldt_info.base_addr == 0 && ldt_info.limit == 0 &&
280 (func == 1 || LDT_empty(&ldt_info))) {
281 ldt_p->a = 0;
282 ldt_p->b = 0;
284 else{
285 if (func == 1)
286 ldt_info.useable = 0;
287 ldt_p->a = LDT_entry_a(&ldt_info);
288 ldt_p->b = LDT_entry_b(&ldt_info);
290 err = 0;
292 out_unlock:
293 mutex_unlock(&ldt->lock);
294 out:
295 return err;
298 static long do_modify_ldt_skas(int func, void __user *ptr,
299 unsigned long bytecount)
301 int ret = -ENOSYS;
303 switch (func) {
304 case 0:
305 ret = read_ldt(ptr, bytecount);
306 break;
307 case 1:
308 case 0x11:
309 ret = write_ldt(ptr, bytecount, func);
310 break;
311 case 2:
312 ret = read_default_ldt(ptr, bytecount);
313 break;
315 return ret;
318 static DEFINE_SPINLOCK(host_ldt_lock);
319 static short dummy_list[9] = {0, -1};
320 static short * host_ldt_entries = NULL;
322 static void ldt_get_host_info(void)
324 long ret;
325 struct ldt_entry * ldt;
326 short *tmp;
327 int i, size, k, order;
329 spin_lock(&host_ldt_lock);
331 if (host_ldt_entries != NULL) {
332 spin_unlock(&host_ldt_lock);
333 return;
335 host_ldt_entries = dummy_list+1;
337 spin_unlock(&host_ldt_lock);
339 for (i = LDT_PAGES_MAX-1, order=0; i; i>>=1, order++)
342 ldt = (struct ldt_entry *)
343 __get_free_pages(GFP_KERNEL|__GFP_ZERO, order);
344 if (ldt == NULL) {
345 printk(KERN_ERR "ldt_get_host_info: couldn't allocate buffer "
346 "for host ldt\n");
347 return;
350 ret = modify_ldt(0, ldt, (1<<order)*PAGE_SIZE);
351 if (ret < 0) {
352 printk(KERN_ERR "ldt_get_host_info: couldn't read host ldt\n");
353 goto out_free;
355 if (ret == 0) {
356 /* default_ldt is active, simply write an empty entry 0 */
357 host_ldt_entries = dummy_list;
358 goto out_free;
361 for (i=0, size=0; i<ret/LDT_ENTRY_SIZE; i++) {
362 if (ldt[i].a != 0 || ldt[i].b != 0)
363 size++;
366 if (size < ARRAY_SIZE(dummy_list))
367 host_ldt_entries = dummy_list;
368 else {
369 size = (size + 1) * sizeof(dummy_list[0]);
370 tmp = kmalloc(size, GFP_KERNEL);
371 if (tmp == NULL) {
372 printk(KERN_ERR "ldt_get_host_info: couldn't allocate "
373 "host ldt list\n");
374 goto out_free;
376 host_ldt_entries = tmp;
379 for (i=0, k=0; i<ret/LDT_ENTRY_SIZE; i++) {
380 if (ldt[i].a != 0 || ldt[i].b != 0)
381 host_ldt_entries[k++] = i;
383 host_ldt_entries[k] = -1;
385 out_free:
386 free_pages((unsigned long)ldt, order);
389 long init_new_ldt(struct mm_context *new_mm, struct mm_context *from_mm)
391 struct user_desc desc;
392 short * num_p;
393 int i;
394 long page, err=0;
395 void *addr = NULL;
396 struct proc_mm_op copy;
399 if (!ptrace_ldt)
400 mutex_init(&new_mm->ldt.lock);
402 if (!from_mm) {
403 memset(&desc, 0, sizeof(desc));
405 * We have to initialize a clean ldt.
407 if (proc_mm) {
409 * If the new mm was created using proc_mm, host's
410 * default-ldt currently is assigned, which normally
411 * contains the call-gates for lcall7 and lcall27.
412 * To remove these gates, we simply write an empty
413 * entry as number 0 to the host.
415 err = write_ldt_entry(&new_mm->id, 1, &desc, &addr, 1);
417 else{
419 * Now we try to retrieve info about the ldt, we
420 * inherited from the host. All ldt-entries found
421 * will be reset in the following loop
423 ldt_get_host_info();
424 for (num_p=host_ldt_entries; *num_p != -1; num_p++) {
425 desc.entry_number = *num_p;
426 err = write_ldt_entry(&new_mm->id, 1, &desc,
427 &addr, *(num_p + 1) == -1);
428 if (err)
429 break;
432 new_mm->ldt.entry_count = 0;
434 goto out;
437 if (proc_mm) {
439 * We have a valid from_mm, so we now have to copy the LDT of
440 * from_mm to new_mm, because using proc_mm an new mm with
441 * an empty/default LDT was created in new_mm()
443 copy = ((struct proc_mm_op) { .op = MM_COPY_SEGMENTS,
444 .u =
445 { .copy_segments =
446 from_mm->id.u.mm_fd } } );
447 i = os_write_file(new_mm->id.u.mm_fd, &copy, sizeof(copy));
448 if (i != sizeof(copy))
449 printk(KERN_ERR "new_mm : /proc/mm copy_segments "
450 "failed, err = %d\n", -i);
453 if (!ptrace_ldt) {
455 * Our local LDT is used to supply the data for
456 * modify_ldt(READLDT), if PTRACE_LDT isn't available,
457 * i.e., we have to use the stub for modify_ldt, which
458 * can't handle the big read buffer of up to 64kB.
460 mutex_lock(&from_mm->ldt.lock);
461 if (from_mm->ldt.entry_count <= LDT_DIRECT_ENTRIES)
462 memcpy(new_mm->ldt.u.entries, from_mm->ldt.u.entries,
463 sizeof(new_mm->ldt.u.entries));
464 else {
465 i = from_mm->ldt.entry_count / LDT_ENTRIES_PER_PAGE;
466 while (i-->0) {
467 page = __get_free_page(GFP_KERNEL|__GFP_ZERO);
468 if (!page) {
469 err = -ENOMEM;
470 break;
472 new_mm->ldt.u.pages[i] =
473 (struct ldt_entry *) page;
474 memcpy(new_mm->ldt.u.pages[i],
475 from_mm->ldt.u.pages[i], PAGE_SIZE);
478 new_mm->ldt.entry_count = from_mm->ldt.entry_count;
479 mutex_unlock(&from_mm->ldt.lock);
482 out:
483 return err;
487 void free_ldt(struct mm_context *mm)
489 int i;
491 if (!ptrace_ldt && mm->ldt.entry_count > LDT_DIRECT_ENTRIES) {
492 i = mm->ldt.entry_count / LDT_ENTRIES_PER_PAGE;
493 while (i-- > 0)
494 free_page((long) mm->ldt.u.pages[i]);
496 mm->ldt.entry_count = 0;
499 int sys_modify_ldt(int func, void __user *ptr, unsigned long bytecount)
501 return do_modify_ldt_skas(func, ptr, bytecount);